Changelog

Restore deleted content from the trash — deleted drafts land in a trash and stay restorable for 30 days, with the name of whoever deleted them, so clearing out forty drafts and changing your mind on eight is one action.

Track AI Overview coverage over time — the share of your keywords where Google cites you, month after month next to your positions, with a Content Score on the keywords that show an AI Overview without citing you. Available from 21 September, once the AI Overview data has finished computing.

Compare your AI Overview citations with your competitors — one chart shows who gets cited on your keyword universe, brand by brand, and clicking a competitor who is ahead opens the keywords carrying the gap. Available from 21 September, once the AI Overview data has finished computing.

Find the keywords where competitors are cited in AI Overviews and you are not — the keywords a competitor owns in the AI Overview while you are absent, ranked by volume and ready to add to the Planning. Available from 21 September, once the AI Overview data has finished computing.

Positions back to seconds — the Positions tab and its charts load without timing out, including on the largest workspaces over a 365-day range with a year-over-year comparison.

AI Assistant replaces the AI Search channel — the channel comes straight from GA4's native channel group, Claude joins ChatGPT, Perplexity, Gemini and Copilot as a tracked engine, and unrecognised sources land in "Other" instead of being dropped.

Manage your own brand aliases — aliases can be added, edited and deleted from the Brand screen without going through support. Adding competitors stays with the Semji team.

AI Overview data over MCP — Performance, Pages and Planning expose their AI Overview data, and the GEO focus prompt can be read, set and analysed from your AI assistant.

Latest generation of models for body copy and Ask AI+ — generated content no longer contains em dashes.

The Positions screen loads again on a 365-day range with a year-over-year comparison, instead of hanging and returning a network error.

The folder filter "is not" no longer silently drops every content without a folder, so the content and click totals are correct again.

Hiding a channel in a chart legend keeps that channel hidden when you change period.

AI Overview previews show their real content and their cited sources again, instead of empty bullets and opaque Google links.

The link to the associated pages opens the list already filtered.

A keyword filter stays applied when you switch between Drafts and Published Content.

Launching Atomic Content on a bulk selection no longer crashes the whole page.

The folder selector on content creation shows the real folder instead of "General" while the list loads.

The AI Usage workspace rankings show workspace names instead of technical identifiers, and only the workspaces you can access.

The selection counter no longer freezes after "Select all", so the cap can be cleared.

Pasting several existing URLs on draft creation works again.

Created drafts carry a last modified date.

The AI Overview chart and the table below it share the same filter.

The publications chart compares the right periods again.

The Focus keyword positions chart no longer counts unpublished pages.

The chart shows a loading state instead of freezing when you switch view.

The HTTP badge no longer shows 301 on every redirected page.

Launching a report that is already running now says so explicitly.

Competitor reports are no longer blocked by a page count validation, and their SERP is refreshed every month.

The prompt metrics load in seconds with a filter applied.

The Agents menu is reachable from the Agents and Activities tabs, and the "Run by" column is filled on older activities.

A Workspace agent can no longer be saved without a channel, and the agent configuration form no longer breaks when the text area is resized.

The wording no longer refers to "Search Intelligence".

Atomic Content no longer errors on a double click on Continue, and the sidepanel button and the assignee avatar display correctly.

Sorting by name respects ascending and descending order, the workspace name is saved, and "Send to CMS" has a default.

The Content Hub search resets when you change tab, and an explicit error is shown for URLs without workspace access.

New Competitive Intelligence agent — it runs every month alongside the Intel Hub SEO report, detects the competitors that changed course (new page clusters, local deployments, editorial pushes on topics where you are absent) and posts a structured report in Slack or Teams with an example keyword, a recommendation and a link to the matching Intel Hub view.

Custom Workspace agents — an agent with access to your whole workspace analyses Intel Hub data, adds drafts to the Planning and publishes what it did in Slack or Teams, on a daily, weekly or monthly schedule. Some tools are still missing for Workspace agents to reach parity with MCP, and are being added progressively.

Semji MCP v2 — the server doubles its number of tools, covering the SEO Data Explorer, GEO prompts and sources with their recommendations, Pages and Performance, Content Ideas searches, Knowledge Base writes, editor comments, user listing and invitations, and agent launches on a batch of contents. Every toolset is enabled by default, with a read-only mode and per-toolset scoping, and the Semji connector is listed in the official ChatGPT and Claude connector libraries for one-click install.

Saved views on every table screen — Intel Hub SEO and GEO, Pages, Planning, Content Ideas and Performance save your filters, columns, sort order and display mode in a named view, reopened in one click from the chips at the top of the screen and shareable with the whole workspace. Default views ship on every screen, on current offers only.

AI Overview column in the Positions tab — a filterable column qualifies each tracked position as no AI Overview, AI Overview present and cited, or AI Overview present and not cited, which is the missing explanation behind a traffic drop on stable rankings. Positions tab only in this first version.

Content Ideas in the public API — launch a search on a keyword, check its status, list its results with all their data and the same filters as the app, and delete a search. Editing a single idea is not exposed in this version.

The Integrations screen layout has been improved.

AI+ runs on the latest generation of Claude models — the outline, title, introduction, key points and FAQ steps all moved.

Retrying a title or meta description in Ask AI+ proposes something new — instead of a near copy of the previous try.

A text selection inside a link can now be commented.

AI assistant traffic from ChatGPT, Gemini and Perplexity feeds the AI Search channel again after Google's new "AI Assistant" channel group in GA4.

Agents fill an empty title and meta description reliably, whichever way they write, so bulk product sheets no longer come back half done.

External MCP servers connected to an agent are available again on the latest Claude models.

An unreachable external MCP server no longer fails the whole agent run.

The Content Score no longer disappears from the Planning after a focus keyword change.

Related keyword suggestions no longer fail when a keyword has no search volume.

The focus keyword field on content creation shows the full keyword and lets you place the cursor and edit it.

The folder lists on the content creation grids are complete from the start, on every row.

On Greek and Bulgarian workspaces, competitors are detected in Search intents and search volumes are populated.

SEO analyses no longer fail when Google serves an anti-bot page in place of results.

AI Overview sources show their real URL and domain instead of a Google redirect link.

Visibility, mention and citation rates can no longer turn negative when a day's analysis finishes late.

The AI Usage rankings show user names instead of identifiers, and the User filter is populated.

A generation interrupted by an infrastructure restart no longer stays "in progress" forever, and its credits are refunded.

Custom instructions accept up to 5000 characters, the same as the platform.

Keyword positions no longer error on a single-day date range, and the custom date mode can be cleared again.

Ask AI+ template usage shows a readable label in the credit history.

The "Mentioned brand" filter of the Prompts Explorer no longer keeps prompts without a real mention.

The prompts list with a mentioned or cited filter loads in seconds instead of timing out.

SEO analyses no longer fail on long search intent lists.

Launching an AI+ generation in a predefined block no longer erases its content.

The FAQ template custom instruction in Ask AI+ is no longer rendered as markdown.

The fact-checking filter dropdown displays correctly, and version labels share the same font weight.

The duplicated Name field is gone from the custom agent forms.

Writes that reference a folder no longer fail with a permission error.

The workspace dropdown displays correctly when adding a user to an organization.

Creating a workspace no longer crashes.

AI Overviews go live in France — Google has started rolling out AI Overviews in France and AI Overview coverage is switched on across all French workspaces, so a French keyword that triggers one gets GEO recommendations. They mostly trigger on informational and question-type keywords, and coverage grows as the rollout progresses, which is also why a keyword can show an AI Overview to a signed-in Google user and none in private browsing.

AI Usage dashboard — an organization-level view of the content published and the hours saved next to the credits consumed, broken down over time and by workspace, user and feature, so a manager can see who drives adoption.

DeepSeek coverage — DeepSeek joins the AI engines tracked in Intel Hub GEO, so brands competing on the Chinese market can monitor their AI visibility there. It tracks brand mentions only; source citations are not available for this engine.

The Sources view is easier to read — a new Prompts column shows how many of your tracked prompts cite each source, top domains are aggregated by root domain rather than one line per subdomain, and percentage trends display in points.

Suggested prompts can be filtered by keyword.

Every content agent runs on the latest AI models — for more reliable generation across the editor and the Agents.

Synchronize every page of a workspace in one action — the 100-page batch limit is gone, so a large workspace no longer needs dozens of manual syncs.

Generating with Ask AI+ on a selection ending in a table cell no longer opens a stray side panel or writes outside the editor, and the result is inserted normally.

The clicks table respects the web and Discover channel filter instead of summing both channels.

Performance tracking no longer disappears after editing a content.

A platform whose scrape is running late no longer disappears from the "Visibility score by platform" chart.

Filters applied on the SEO side are no longer silently carried over when navigating to GEO.

A competitor set up as a folder no longer overwrites its parent domain's bar in the report charts.

SEO and GEO exports use distinct file names.

Editing the name or site of "My brand" in the AI Context is saved correctly.

Creating a draft to update an existing URL launches the analysis automatically again.

The "Sync with your calendar" button opens again.

A Google OAuth error no longer breaks Search Console metrics and top-keyword data.

Content analyses no longer fail by timeout on the new analysis pipeline.

AI+ content generation no longer adds unwanted bold to paragraphs.

Multi-organization users no longer hit a 403 when calling the public API.

Opening the comments tab from a notification email works.

The Basic plan no longer shows unlimited credits.

CMS Publisher agent — it takes a finished piece all the way to WordPress, Strapi, Contentful or Drupal, learning how your existing articles are built (their content blocks and their structure) and reproducing that layout when it publishes. It has memory, so a correction you give it once holds for the next publication.

A new generation of agents — Content Master, EEAT Editor, Product Fact-checker, the GEO expert and your Custom agents all learn from your feedback and remember it, ground their answers in your Knowledge Base, write in your brand voice (both on by default), and lay out a plan before working through a complex task. No reconfiguration needed.

Connect your own tools through custom MCP connectors — add a connector once and your agents can pull data from it and act on it inside a conversation: Meta Ads to turn winning paid campaigns into organic content, your PIM for specifications, attributes and pricing, Screaming Frog to factor in broken links, missing titles and thin pages, or a document workspace to read briefs and guidelines and write output back.

Fact-checking tags no longer leak into the HTML agents produce or that you copy into your CMS.

The trash icon in the AI settings deletes file and URL sources again.

Bulk optimization applies to the current draft instead of an older published version.

The "Sync with your calendar" button opens again.

An SEO analysis no longer fails when one competitor page is oversized.

The delete-agent modal closes properly, opening a conversation from a deleted agent no longer errors, and the crawl tool returns content reliably.

Atomic Content retry works again, and the plus sign in related terms is kept on extraction, for example T+1.

Clearer "incomplete configuration" message, aligned metric columns in the Prompts grid, steadier loading states, and a fixed search-volume mismatch between prompt selection and the editor.

Minor layout fixes on the overview and the collapsed side menu.

Proactive agents in Microsoft Teams — a new Workspace agent type runs on a schedule and posts its results straight into a Teams channel, starting with the Content Opportunity agent, so the team sees what to write next without opening Semji. Workspace agents cover Planning and the Intelligence Hub, and the Semji app is on the Microsoft marketplace.

Agents search your Knowledge Base on their own — an agent pulls product documentation, internal references and company definitions in context while it works, with nothing to paste in by hand. Activated on 25 June 2026.

Sources Explorer — a Sources screen shows which web sources the AI engines actually cite on your topics, with totals for sources, unique domains and your own pages, filters by AI platform and page type, and a spreadsheet export.

Automatic prompt suggestions — Semji proposes the prompts worth monitoring from your website content, classifies them by topic, and lets you filter, search and add them in bulk, so a new GEO workspace is set up in minutes.

Custom Agents take 100K characters of instructions — the limit doubles from 50K, so detailed editorial guidelines fit in one agent.

Inviting a user from Settings > Organization > Users no longer fails with a "Something went wrong" error.

CSV files exported from Semji can be re-imported again instead of being wrongly rejected.

Drafts created by a bulk import on existing pages all show up in the Planning "Drafts" list, including older imports.

Links pointing to a PDF no longer stay stuck loading in the Knowledge Base; they import correctly or fail clearly.

SEO analysis runs reliably right after a focus keyword change, with no page reload.

The "locate in editor" and "apply suggestion" actions work on agent suggestions that follow a quote or testimonial block.

Re-adding a competitor attaches its data to the dashboard again.

The AI Visibility chart respects the Brands filter.

The Prompts data grid no longer hides evolutions or shows wrong deltas, and numeric columns line up.

AI-search analysis no longer fails when an engine returns a refusal, and language detection is back.

Dashboard widgets no longer stay stuck loading when there is no data yet.

The page title shown is the workspace page's rather than a competitor's, and the workspace favicon displays correctly.

Source links in the "Usage examples" modal for related terms are clickable again, and a stray URL no longer appears outside the link-edit modal.

HTML tags typed in a comment no longer disappear, and bold renders correctly on Greek labels.

Workspaces with no Search Console connected see the "Connect Search Console" prompt instead of a misleading empty state.

A 308 redirect on URL import is no longer mislabelled as a 301.

Sorting by "Last published" works reliably again, and "Manage goals" is back in the Analytics view.

In the collapsed side menu, sub-menus are reachable again without opening the wrong one.

AI Visibility Score — a single 0-100 KPI for your brand's presence across ChatGPT, Gemini, Claude and AI Overviews, with its trend, a competitor benchmark, a per-platform breakdown and an export, from the dashboard down to each individual prompt.

Semji MCP server — connect an AI assistant to your workspace over the Model Context Protocol, scoped to one organization with OAuth, to plan the editorial calendar, bulk-generate Atomic Content drafts, import pages with their focus keyword, run SEO and GEO analyses and publish, then review the drafts in Semji. The reference is on developers.semji.com.

Agents no longer get stuck on "Thinking" in production; runs complete normally again.

The writing agent no longer finishes without writing the article into the editor on long content.

Atomic Content no longer shows an empty result while marked done; generation fails clearly instead of saving blank content.

"No AI Overview found" is no longer shown when an AI Overview actually exists in the SERP, and Semji re-checks to detect it reliably.

SERP competitor insights display on German-interface workspaces instead of an empty block.

Excel uploads to the Knowledge Base no longer stay stuck loading; large files are processed reliably or fail clearly.

Drafts created by a bulk import on existing pages appear correctly in the Planning "Drafts" list.

The GEO tab is reachable again after configuring a GEO Expert agent, instead of redirecting to SEO.

The editor "back" button is no longer slow.

"Already in planning" and "already a page" badges are clickable, and jump to the existing content.

AI-search responses record their sources and rankings on self-citations.

Fixed an error in the ChatGPT analysis rendering, and a truncation on Top Prompts question generation.

Competitor traffic normalization is accurate for workspaces scoped to a subfolder.

The AI properties funnel-stage panel no longer flickers while scrolling, and the Notifications screen displays correctly.

Content assignment no longer sends an incorrect assignee value.

Public API V1 and the developer portal are generally available — create and update drafts, mark them published, launch keyword and prompt analyses and pull back the recommendations and content score, trigger Atomic Content generation, import and refresh pages, list credit consumption and manage workspaces, folders, members and statuses. The reference, quick-start guide and code samples for every endpoint are on developers.semji.com. Intelligence Hub, Content Ideas, Performance and launching AI agents are not in V1.

Gemini, Claude and Perplexity join ChatGPT and AI Overviews — mentions and citations are now collected from five AI engines, and engines outside your plan can be added.

Agents no longer end as completed with no output when the model invents a tool name.

Content Localizer, From Scratch and the other content-editing agents no longer silently drop their edits on an unexpected model output.

Long-running agents no longer fail after a few minutes on an infrastructure error, so runs are not lost and credits are not spent for nothing.

Fact-checking no longer shows raw ** markers or drops parts of facts, including inside table cells.

Opening content in Atomic review mode no longer crashes.

No more slowdowns after several minutes of writing.

Occurrence counters and usage examples for related terms work on Greek and Bulgarian workspaces.

Version history shows the synced content score after a website sync, not the old one.

The before/after magnifying glass targets the right phrase.

Limited Guest Writer users can create drafts from existing content again.

Daily AI scraping no longer leaves prompt responses stuck in "Pending" after a transient infrastructure issue.

The cancel and continue buttons on Atomic generation and review are large enough to click comfortably.

The agents chat disclaimer no longer breaks in the middle of a word.

A competitor page's H1 is readable in full instead of truncated in the column.

The Ask AI template list stays visible while you edit a template.

Saving a prompt as a template pre-fills the description with the prompt instructions, and version comparison lists versions in chronological order.

The custom agent appears at the top of the agents list, and outbound links come before inbound links in the internal-linking sidebar.

User access toggles are aligned again on the workspace users screen.

The import loading indicator no longer stays stuck after the import has completed.

Competitor Visibility — the Prompts table shows a mentions ranking and a citations ranking per prompt, and each AI response lists which competitors are mentioned and cited, in the order they appear.

Bulk and inline editing of prompts — select one, several or all prompts to bulk-edit topic, funnel stage and persona, bulk-export or bulk-delete with a typed confirmation, and edit or delete a single prompt from its row menu.

Prompts export — export the Prompts table to Excel or CSV in one click with every active filter preserved; large exports arrive by email.

Content Localizer handles regions and local sources properly — a reworked prompt brings native-quality writing, correct localization of local institutions, regulators and statistics instead of foreign authorities left in a translated article, and per-locale typography for decimal separators, currency placement and date formats.

Legal Compliance checks regulated facts on official sources — on demand, the agent confirms a regulated fact online against government domains, regulators, Légifrance and EUR-Lex, and nothing else.

Agents stop reporting work they did not do — a new behaviour layer blocks silent failures, false "done" messages and invented facts, with stricter rules again on Legal Compliance.

Bulk agents and Atomic Content survive large batches — fewer execution errors when agents and Atomic Content run in bulk.

GEO prompt consumption in Billing — admins see a live count of prompts used against the plan limit in Billing & Invoicing.

Changing the content no longer redirects to the previous optimization.

Ask AI advanced mode no longer truncates the instructions box.

The publication date field shows on first connection.

Export scheduling fields reflect the real schedule, and competitor top-pages exports work on large catalogues.

Search-intent analyses no longer fail on heavy workloads.

Brand ranking in AI response details is correct.

Prompts can be sorted by persona and funnel stage.

The latest-responses table and the extracts panel are lighter to read.

Mentions and citations ranking tooltips show rank and rate side by side.

The # prefix is gone from the rank columns in competitor rankings by topic.

The brand-excluded filter is back on the traffic chart.

The Top 3 agent no longer shows a duplicated Name field.

Bulk status changes no longer fail silently.

Draft titles are sized more sensibly.

Leaving with unsaved changes shows the confirmation popup again.

Fixed a rare crash when the model was overloaded.

The "set as default" button on AI+ presets stays visible while scrolling.

The bulk delete button is no longer wrongly disabled.

Links in related-term extracts are clickable again.

The usage-example modal shows the related term you clicked rather than the focus keyword.

Versions appear in a sensible order in the history panel.

H4 matches paragraph size, so the heading hierarchy reads correctly.

Fixed an icon size glitch in the organization creation flow.

Labels and icons line up in the sidebar menu.

Track your brand in AI answers, prompt by prompt — daily monitoring across ChatGPT, Perplexity, Claude and the other main engines shows how often your brand is mentioned against the competition on the prompts that matter to your market, with a drill-down into every individual AI answer to see who is quoted and a filter by your own brand or a given competitor. On the GEO plan.

Four new agents in the library — Content Master, a conversational SEO and GEO writer that proposes work based on the draft and adapts its output format; EEAT Auditor, with YMYL classification, a top 3 SERP benchmark and before/after suggestions; Editorial Assistant Top 3, a comparative analysis against the top 3 SERP competitors; and Product Page Specialist, covering HTML template reading, semantic optimisation of limited editorial space and regulatory fact-checking.

Bulk execution on all four new agents — run them on several drafts at once from the Planning. EEAT Auditor and Product Page Specialist write their before/after rewrites straight into the draft in bulk-with-editing mode; Editorial Assistant Top 3 stays display-only by design.

AI credit indicators are gone — the counters are removed across the platform, so teams run AI workflows without watching a number.

Drafts imported from CSV no longer show an empty initial version in the history panel.

SEO analysis no longer loops indefinitely when a scraped competitor heading contains an emoji or a special character.

The GEO Expert agent no longer fails silently when analysing content with many large external sources.

Sorting on the last-modified column works.

Bulk CSV imports no longer leave orphaned content versions behind.

The top keyword report no longer surfaces the wrong report from an optimistic identifier.

Generations no longer stay stuck after a deployment.

Agents no longer stay in a reasoning state after finishing their work.

The editor step no longer fails on leftover thinking blocks.

Social Repurposer no longer blocks in bulk mode.

Agent Showcase — a Discover screen is the default entry point for agents: browse every template, read what each one does, and configure it yourself.

Two new built-in agents — Fact-Checker verifies claims against sources automatically, and Editorial Assistant Top 3 now runs in auto mode.

Agent Hub rebuilt — tab-based navigation, one unified Activities screen with clearer columns, updated visuals for every template, and a simpler configuration for editor agents.

Ask AI is twice as fast — average wait times halved since early April, and generations over thirty seconds are four times rarer. Simple queries are close to instant, and long articles with rich context gain the most.

A better Strategist for Atomic Deep and Optimize — stronger output on FAQ-type questions, with related terms worked in properly.

New navigation menu — the updated menu ships with a dedicated Agents entry for reaching agent configuration quickly.

Isolate one series in an Intel Hub chart — Ctrl-click or Cmd-click a legend item to show that series on its own.

A page can be deleted again during keyword selection in the content creation flow.

Bulk imports no longer return a 403.

The Matomo token is no longer sent as a GET parameter, which some customer setups blocked.

Session and bounce-rate trend curves display correctly.

Clicking "Details" on a KPI no longer disconnects the user on legacy workspaces.

Untranslated English text and typos corrected on the Performance screen.

Clicking a row opens the drill-down instead of redirecting to the report.

The black tooltip is readable.

GEO related terms no longer show English extracts or duplicate sources.

Bold keywords are back on German questions.

Fixed the widget height on traffic and pages trends.

The web extension works again.

Wording corrected on the GEO draft, the Planning filters and the Pages section.

Date picker labels fit on one line.

Atomic Deep and Flash apply the GEO recommendations — generated content takes the ChatGPT and AI Overview recommendations into account alongside classic SEO, automatically and with no setup.

Launch agents in bulk — run an agent on several contents at once.

Ask AI+ answers three times faster — a new model also follows the instructions more closely on requested length, format and number of items.

Fact-checking raises fewer false alarms — it recognises when the information already exists in the customer's documents or in the search results, instead of flagging it for review.

Knowledge search picks better sources — when the custom instructions ask for a given kind of source, such as government sites, the search targets them, and thin sources like empty product sheets, directories and classified ads are set aside.

A unified activities listing in the Agent Hub — one set of columns across agent activity.

Better Norwegian localization, and the Content Localizer warns when search intent does not match between languages.

Competitor data calls retry automatically on a transient error, and AI properties use newer models for more accurate results.

Bulk import V2 ships with its corrections, and exporting content is far faster on large volumes.

GEO citation tooltips name their source, ChatGPT or AI Overview, and alert components are consistent across the app.

Renaming an AI property option no longer throws an error.

The brand-excluded filters work correctly again, on the drill-down and on brand classification.

Published content lands in the right folder.

Competitor reports are no longer blocked after creation, and reports awaiting approval are no longer wrongly marked as failed.

Fixed the wrong icon on the Traffic tab.

Fixed special characters in AI-generated lists.

The tooltip on custom agent configuration is readable.

Repaired 33 broken help links across the app, plus the "allow semjibot" link.

Fixed a broken icon in the organization creation flow.

The page deletion modal is clearer.

Overview counters no longer load forever.

A new model generates titles — better quality on the TITLE tag.

A better Strategist for Atomic Deep and Optimize — higher-quality output, with related terms worked in properly.

Atomic Content is out of beta — the Beta label is gone.

More relevant related terms — the SEO recommendations come from a new model.

Help links point at the current documentation across AI+ Agents.

AI presets save more smoothly and detect changes properly.

Competitor favicons and heatmap cells are clickable, and open in a new tab.

Atomic Deep Research no longer generates empty content.

Ask AI+ no longer fails to generate on long content.

The due date survives adding a page to the Planning from Pages.

The date picker displays correctly again.

Search and replace works in tables, in bulleted lists and when deleting text.

AI-generated bulleted lists display correctly.

The "add another source" link works again.

Fixed agreement errors in the notification messages.

Atomic Content with the Knowledge Base takes 10 minutes instead of 15 — and the app says so when the Knowledge Base is on, since more context means more work.

Fewer failed Atomic runs — the errors that led to cancelled or broken Atomic Content are down, so runs are more reliable.

Better prompts for titles and meta descriptions.

The publication date is on the published content tab.

Number formatting is correct.

Repaired broken links and several translations across the app.

The search bar behaves sensibly.

The last modified date updates.

Sorting by an AI property works again.

Fixed the navigation bug on the Performance panel.

Pick an exact date — configure a specific date rather than choosing from the preset ranges.

An "Improve instruction" button — generate the instruction for a custom agent instead of writing it from scratch.

Excluded terms apply to agent output — terms you exclude stay out of what an agent writes.

The GEO Expert agent covers AI Overview — recommendations for Google's AI Overview alongside the ChatGPT ones.

The conflict modal names who edited last — you see that another editor is active and get a clear action, without being blocked.

AI presets update in place, which makes them quicker to configure.

Going back preserves the previous URL.

Excluded terms are respected in AI generations.

Fixed the broken display of the credit logs, and standardised how agent credits are named.

Fixed scrolling in the outline tab.

AI Overview analysis in AI+ agents — the From Scratch and GEO Expert agents can analyse the AI Overview alongside the SERP.

Turn off web search in AI+ settings — disable the web search source so generation works from your own material only.

A stronger model behind Atomic Optimize — better generation quality.

Reworked strategist, writer and editor steps for Atomic Optimize and Deep, which makes for more relevant drafts and a smoother run.

Intelligence Hub reports run automatically — a report starts on its own rather than waiting for a manual launch.

Scrollbars fade after a second of inactivity in tables.

The focus keyword is in the editor tab title, so several open drafts are tellable apart.

A clearer bulk modal when launching Atomic Content on a selection.

Atomic Optimize no longer blocks while reading the draft, so generation runs through.

The credit limit is handled correctly on Atomic generation.

Analyses are no longer triggered after their expiry.

URLs can be up to 4,000 characters.

The lock badge works again.

Fixed the settings grid layout.

Agents analyse a page's full HTML — an AI+ agent retrieves the complete HTML of a web page, which makes a technical audit of schema.org tags, semantic structure and accessibility possible.

Add a page to the Planning from Performance — one click from a Performance report puts the page in the Planning.

Command bar — Cmd+K or Ctrl+K reaches any page, module or draft: type what you are after and press Enter.

A new icon library across the app, with the technical and UX work behind it.

AI settings are no longer fully locked for users who have access to the AI suite without the "manage AI files" option.

Marking several drafts as published no longer leads to a blank page.

Fixed persona filter sorting in the grid headers.

Special characters are no longer escaped in the From Scratch prompt.

GEO Expert agent — it audits your content and returns targeted, actionable recommendations for showing up in ChatGPT answers, with AI Overview coming.

Recommendations for AI Overview — Semji previews the AI Overview present on a SERP and recommends the related terms to cover, ahead of the feature reaching France.

Set the focus prompt by hand — define the exact search intent you want the content to serve, and Semji tailors its ChatGPT recommendations to that prompt. After the first GEO analysis it also proposes four more prompts to choose from.

Atomic Content Deep — writing from scratch runs through an agentic workflow, which lifts the quality of a first draft.

Agents read more of the SERP — people-also-ask blocks and ads join the top pages, so recommendations rest on a full view of the results.

Show and hide sections in an agent's answer — audits are long, so they are now easier to read through.

Agents read comments — an agent uses a colleague's review comment as an input when improving the content.

A steadier editor — smoother and more robust, with fewer conflict modals and a layout that holds at every screen width.

A new version history — easier to read, and the content score shown matches its version.

Choose your columns — each user sets the order and visibility of table columns across the platform.

Position tracking covers the top 30.

Tables load faster.

GEO data is included in analysis exports for workspaces with access.

Fixed the display of the position filter.

Fixed text replacement in Ask AI+.

A Localization Expert agent template — it takes a market's own context into account when adapting a content, so an English master can be rewritten for the Dutch market rather than translated word for word.

Agents can edit content in the editor — an agent can insert before or after a block, or replace it, instead of returning text for you to paste.

Save your AI settings as presets — the generation options you use most are stored once and picked up again in every draft.

Compare a content before and after an agent has worked on it — the two versions sit side by side with the full modification history.

Point an agent at a specific block — selected text is passed to the agent with its highlight, so the work stays on the passage you chose.

Focus Prompts are better aligned with the search intent behind the keyword.

Search volume and average position appear straight away instead of after a delay.

Exporting the URLs from the Pages tab works again.

Agents can call the SEO and GEO tools — an agent runs the same audits the platform does and writes its recommendations from those results, which makes them specific enough to act on.

Agents can read the Page tab in the editor — the page's own metrics are part of what the agent sees before it answers.

GEO analyses stop instead of looping — when an analysis fails, the interface says why in plain language and you relaunch it, rather than watching it spin.

Competitor analysis can be narrowed to specific parts of a site rather than the whole domain.

The "AI+ Suite" is called "AI+ features" in the settings.

Cancelling an optimization works again, and the progress indicator reports the real state.

Opening the text behind a page from the Pages tab lands on the right content.

The side panel remembers your navigation — moving between recommendations and back no longer resets the panel to its first screen.

Opening the editor from a list is faster.

The generation footer reports the right status.

The bulk settings dialog applies the choices it shows.

A page that is not published and has no URL shows a proper empty state.

An Atomic Content in review with no HTML no longer breaks the screen.

The query parameter for an existing content resolves correctly.

Positions display a real "-" where there is no position, instead of a blank cell.

The Focus keyword's last known position is shown again.

Several interface defects across the app.

The average Content Score persists over time — the chart draws a real history instead of dropping to zero where nothing was stored.

Switching workspace keeps you on Performance — you land on the same screen for the next workspace rather than back at the top of the app. On Legacy plans the screen opens on the Organic Search channel.

Tooltip and chart-title wording corrected across the screen.

An analysis can be relaunched the same day — a second run no longer has to wait until tomorrow.

The paragraph and heading selector is usable on small screens.

Far fewer "An error has occurred" banners — the unexplained red snackbars that appeared during ordinary use have been traced and removed.

A draft assigned to another user opens correctly.

The assignment icon shows the right person.

The page no longer flickers while data loads.

The all-channels tab shows the correct overview figures.

The metadata bar renders correctly.

Importing several URLs at once works.

Filtering on a position beyond 100 returns the right rows.

Multi-channel metric deltas show the right value and suffix.

A channel card with no service connected says so instead of showing an empty metric.

Analytics requests no longer return 403 on multi-channel screens.

Multi-channel filters apply as selected.

The AI Settings error message is correct when the optimized content is in review.

The side panel resizes in the Pages tab.

AI+ template usage is recorded in the version history.

The "Take a tour" button no longer appears on the Brand Voice screen.

Intelligence Hub loads when no AI Properties are configured.

Icon alignment and style overrides corrected in several dialogs and screens.

Pagination survives a page refresh.

Concurrent requests no longer collide and return 409.

Performance covers every channel, not just organic search — the contribution of content is measured against the brand's other channels, which is what turns an argument about the value of SEO into a comparison.

Optimize a batch of contents in one run — bulk Atomic Content is available on request ahead of its general release.

Large workspaces load about ten times faster — the screens that used to stall on the biggest accounts return in seconds.

The Beta label is gone from Atomic Content and Content Ideas — both are now stable features, and the label is reserved for what is genuinely still in development.

Position tracking is accurate again, closing an issue that had run for several months.

The Content Score shown in Planning and in the editor agree for the same content.

Bulk content imports report what happened in clearer terms.

About fifteen display defects across the interface.

German as an interface language — the whole application and the browser extension are available in German.

Multi Channel Performance, first version — content performance is reported across acquisition channels rather than organic search alone, switched on for customers from 16 September 2025.

The extension no longer closes unexpectedly mid-session.

Connecting a service integration no longer triggers an unwanted score recomputation.

An analysis with no keywords returns a clear result instead of an error.

The meta description dropdown opens in the right place.

The folder filter works on the multi-channel view.

The dialog resets its choices correctly, and additional sources are selectable on a bulk run.

Redirected URLs are followed when outgoing links are computed.

The grid header stays in place when it is sticky.

A new menu — it gives back the room the old one took from each screen, holds up on a small display, and switching organization or workspace is now two clicks from anywhere.

Filters survive a change of report date, and a date range can be read as before and after values.

Changing a due date applies immediately and stays applied.

Links are no longer inserted in the Brand Voice and knowledge base editor, where they were never used.

Outline icons use the right colour.

The recommendation side panel is resizable — drag it to the width you work at, which matters most on the screens that carry a lot of recommendations at once.

Filter on "My Position > 100" — the keywords where you are effectively absent, which is the shortest route to a list of content worth creating.

Turkish content is analysed one item at a time. It is slower, and every analysis completes.

Fact-checking uses the other sources you configured, as it was meant to.

Custom goal names are readable in the interface instead of showing their internal key.

Several display and wording defects around the other-sources setting.

The user limit for the extension is reported correctly.

Plan limits display correctly in the extension settings.

A Limited Guest Writer sees the real number of contents, and the full list of folders they have access to.

The meta description displays correctly.

A JavaScript error that could take down the editor is fixed.

The version history no longer shows two conflicting states at once.

A display regression in the reports is corrected, and the competitor's name appears in the report tooltips.

"Already in Planning" is reported accurately when importing pages.

Number filters apply the operator you selected.

Several interface defects across the app.

Optimize an existing content in one click — Atomic Content rewrites what is already published, replacing the parts that have gone out of date, so refreshing an old page costs minutes instead of an afternoon. Available on current offers, not on Legacy.

Five new geolocations: Hong Kong, Singapore, Taiwan, Japan and South Korea — workspaces can be created for those markets.

Tooltip wording corrected.

The right title is shown where the wrong one appeared.

Comparing two versions of a content returns the real differences.

The content conflict dialog appears when it should.

Product tour icons display correctly.

Outlines are built against the competing pages — the generated outline takes the outlines already ranking on the keyword into account, so it covers what the SERP expects rather than what a model guesses.

Titles and meta descriptions read your knowledge and the search intent — generation now uses the knowledge you supplied and the intent behind the SERP, instead of the draft alone.

Copying and pasting a heading keeps its level.

Links behave correctly when clicked and edited.

Starting an optimization shows a loader instead of an apparently idle screen.

The floating AI button sits where it should.

Content prefetch on the page overview works again.

The folder filter is labelled "All content" rather than "All folders", which is what it actually selects.

Applying a filter no longer closes it, and selecting everything no longer closes the secondary filters.

The sidebar displays correctly while editing a brand voice.

Graphs report their loading state accurately.

The funnel stage label is translated.

Grid selection on the dashboard works again.

New geolocation: Greece — workspaces can be created in Greek, with the keyword and position data that goes with that market.

A content on a Legacy plan with no meta description loads correctly.

The browser extension builds and installs again.

A third level of folders — content can be segmented one level deeper, which is what large editorial plans were missing, and folder names are searchable in the selectors rather than scrolled to.

Text Compare reads more clearly on long contents.

The reports render faster.

Data grid headers wrap onto two lines, so a long column name is readable.

Keyword classification on the top pages matches the report's own categories.

Copying a content into a document workspace keeps its formatting.

Page type filtering and its configuration behave consistently.

The keyword length filter takes the right type of value.

The lock icon appears on the Atomic Content button where the feature is not available.

Several interface defects across the app.

AI+ fact-checking on long contents — it covers more of the text, returns a more detailed verdict per claim, and Ask AI+ stays responsive on large drafts where it used to slow down.

Pages loads significantly faster — the gain is largest on the workspaces with the highest page counts, which are the ones that felt it.

Meta descriptions are capped at 140 characters, and copy-paste keeps the formatting it should.

The reports render faster.

The title is optional when creating a content, so a draft can start from a keyword alone.

SAML certificates can be rotated without an interruption.

Further work on the new menu.

The grid resizes correctly when the menu is collapsed.

A page with no title no longer shows an empty row.

The focus keyword analysis no longer reports an invalid date.

Pasting a table keeps its structure.

Deleting from the bulk "mark as published" dialog works.

A failing data refresh no longer retries in a loop.

A "Last modified" column — you can see at a glance which drafts have moved and which have been sitting untouched.

"New content" is recommended where you rank beyond position 100 — Content Ideas and Data Explorer stop suggesting you optimize a page that has no position to improve.

Load time down from eight seconds to two or three — and AI Properties are processed faster, which is what the wait was mostly spent on.

The Content Hub Pages screen loads faster — and keeps its display performance on the workspaces with the most URLs.

The Persona and Funnel Stage columns can be sorted.

The QUESTIONS recommendation prompt was rewritten for relevance.

A batch of interface improvements across the app.

Table cells keep their row and column spans when a content is imported.

Chart colours are consistent between reports.

AI properties filters load, and drilling into one returns the right rows.

The result count no longer reads as undefined.

The Content Score equality filter matches exactly.

The product tour runs to the end.

A malformed model response is handled instead of failing the generation.

The side menu behaves correctly.

Menu item width corrected.

Configure your own AI properties — a workspace owner sets the persona, topic and funnel stage definitions directly, which removes the back and forth that used to make setup the slowest part of an Intelligence Hub rollout. Available from 24 April 2025.

Search intent recommendations are more distinct — near-identical variations are collapsed, capitals and accents are handled properly, and intents are filtered to the workspace's language.

A Brand Voice's style carries through Ask AI+ templates.

Content can be filtered on non-branded and other-brand terms, and reports can be filtered by URL, keyword or title.

Several improvements to how the filters look and behave.

Creating your first Ask AI+ template is a shorter path.

An expired session redirects to the login page instead of failing.

Evolution cells refresh with the rest of the table.

Tooltip and wording corrections across the screen.

The evolution chart is dated correctly.

Connecting a Google Analytics account works again.

The back button in version history returns where it should.

The AI generation selector scrolls properly, and the Ask AI+ dropdown behaves.

Knowledge base errors are explained, and content extraction from knowledge base entries is more reliable.

Custom goal data is included in a page export.

AI access is correct on the smaller offers, and the limited guest role works on Intelligence Hub only workspaces.

Dropdown typography, number formatting, and overflow in the collapsible menu and profile header.

Question recommendations are cleaner and there are more of them — redundant and near-identical questions are removed, spelling and grammar are corrected, every question respects the workspace language, and the list holds up to twenty: five from People Also Ask and fifteen generated.

Font and size corrected for updates shown in the editor's Pages view.

Typography corrected in the input, and the Mind Map view switch was redrawn.

A clearer tooltip on the position evolution metric.

An edit applies immediately and stays applied.

A competitor's top page with an empty SERP is handled instead of failing.

Inserting an optimization saves it.

The grid reloads correctly when you cancel adding a page to the Planning.

The extension's links open the right screen in the app.

Creating a workspace on the Quebec French market no longer returns an error.

The hub menu displays correctly.

A new menu across the app — features and sections are reorganised so both Hubs are visible from anywhere and reaching either takes fewer clicks.

Revenue in your workspace's currency — pick the currency in Workspace > General Settings; existing workspaces were set to their country's currency automatically, so revenue KPIs read in the money the team actually reports in.

Filter on "other brand" traffic — a third traffic type joins branded and non-branded in Content Ideas and the Intelligence Hub, so traffic arriving on a competitor's or a partner's name can be separated out.

Text Compare shows only what changed — the text view lists added and removed terms rather than the whole paragraph, and the coloured background is gone so only the text itself is coloured. On current offers, not Legacy.

Adding a highlighted comment no longer removes the text it was anchored to.

The term detector counts occurrences correctly when stop words are involved.

The conflict modal no longer appears while saving content that is not in conflict.

The date picker works.

An analysis no longer fails under CPU load.

The "tracking limit reached" message is shown only when the limit really is reached.

The count of pages to analyse is correct.

Several interface defects are corrected.

The "Due Date" column header is back in the listing.

The Content Score filter applies.

Generating Atomic Content in bulk no longer errors partway through.

Adding a comment no longer toggles the sidebar.

Preview your AI+ files in place — the files you have given AI+ can be opened and read from the interface, so you can check which document you uploaded without downloading it again.

Navigating between versions with the outline open no longer crashes the editor.

Saving content no longer strips spaces.

Inserting a table works.

The title and meta description alerts say which device they concern.

The generation modal no longer leaves a large gap above the content.

The publication date is shown in the right colour.

The wording and the figures agree.

Several interface defects on webhooks are corrected.

The Manage AI Files tooltip uses the right typography.

Adding a user to an organisation no longer crashes.

Signing up and signing in with Google works again.

The keyword limit message is accurate.

Links in emails point at the app.

Configure a workspace on its root domain — set the root domain and the Intelligence Hub collects the URLs of every associated subdomain automatically, which is the complete picture for a brand whose traffic is spread across several of them.

Clearer filters and columns — Persona filter items are capitalised and the estimated traffic column is labelled "Est. Traffic", so the header fits without truncating.

Sorting holds its state, including on the "my position" column.

A webhook fires when content is marked as published — connect Semji to an automation tool and let publication trigger the rest of the chain, instead of repeating the same manual steps after every article.

Correct a page type by hand — the automatic classification can be overridden on one page or in bulk, so the categories your analyses are grouped by are the ones you chose.

Two new page types, Local and Video — pages that are neither articles nor products can be classified as what they are, and measured as a format.

Generate the missing paragraphs and key points in one go — from an existing outline and a first written section, the rest of the article's AI paragraphs and key points are produced in bulk rather than block by block.

Classifications are correct.

Routing between screens works.

Optimizing an existing page from the Hub no longer fails.

Reports on branded keywords are accurate.

Pipeline notifications are sent.

The traffic filter applies.

The monthly top page reference is retrieved.

The Knowledge Base no longer times out.

Ask AI+ renders and formats its answers correctly.

Fact check results render.

Inserting content no longer leaves an empty line behind, and insertion itself works.

Headings render correctly in the content plan.

Activating and deactivating a brand voice works.

Removing an idea from the planning works.

AI+ Templates — write your AI instructions once, save them as a template and apply them to any content, instead of retyping the same brief on every generation.

Search Intelligence becomes the Intelligence Hub — the name matches what the module does and how the offers are now organised.

Graphs show twelve months rather than thirteen — the extra partial month is gone, so period comparisons line up.

A loader on SERP Insight — the panel says it is working instead of looking empty.

The Ask AI and comment tooltips are positioned correctly.

Switching versions in the content history no longer crashes.

The title and meta description alerts no longer truncate their text.

Trends display correctly.

Column widths in Report Positions are usable.

Navigating between reports no longer shows the previous report's data.

The new draft button responds to a single click, the form is usable on a narrow screen, and an error during creation no longer locks the screen.

Competitor computation respects its limit, the saved configuration keeps the right competitor, and ranked URL positions are computed correctly.

Markdown is parsed correctly during a bulk run, the automatic retry sends the right payload, and a threshold error says what actually happened.

Folder edit modals and status selectors are styled correctly.

Translations are corrected across the app, including the dashboard.

A repeated call for remaining AI units no longer runs on every screen.

Comparing grid columns no longer closes open dropdowns.