How to run a GEO prompt audit?

A GEO prompt audit lets you measure your visibility on generative engines and find the levers that will improve it. Are you cited by ChatGPT, Perplexity, or Google AI Overview? In which position do you appear? Which prompts trigger mentions or citations?

This GEO audit method helps you measure your presence, compare your results with competitors, and adapt your content strategy.

Here is how to run a complete prompt audit so you become a reference for AI assistants.

What are the goals of a prompt audit?

A GEO prompt audit lets you evaluate your brand’s current presence on the new AI models, identify your strengths and weaknesses, and build a matching action plan. It is one of the best practices for improving your visibility on generative search engines.

Identify your key prompts

The first goal of a visibility audit is to define the queries on which you want to appear. The idea is to strengthen your digital presence using the basics of organic search.

Which LLMs do you want to track your prompts on? ChatGPT, Perplexity, Gemini, Claude, or Google AI Overview? Each assistant has its own habits and favorite sources.

Your business goals also shape the choice of prompts: are you looking for traffic, awareness, or conversion? An informational prompt builds authority and qualified traffic. A transactional prompt aims directly at conversion.

Criteria for selecting key prompts:

Identify your targets: bottom of funnel (BOF), middle of funnel (MOF), or top of funnel (TOF);
Define the products or services you want to be visible for;
Prioritize prompts by conversion potential or volume;
Analyze the prompts your prospects use in their research process.

This extra thinking step keeps you from spreading your effort too thin. An effective audit rests on a coherent set of prompts aligned with your goals and your domain expertise.

Visibility assessment and action prioritization

Measuring your digital presence is the second key step. The goal is to understand where you stand so you can decide where you want to go.

This assessment covers several dimensions: are you absent, merely mentioned, or actually cited with a link?

Questions to ask when you evaluate visibility:

On which prompts are we visible, or completely absent?
On which AI engines do we appear: ChatGPT, Perplexity, Claude, or Google Gemini?
In which position do we appear in the citations?
What message is associated with our brand in the generated answers?
Which pages of our site are cited as references?

This comparative analysis reveals your performance gaps across LLMs. You may excel on Perplexity and still be invisible in ChatGPT results. Those facts set your optimization priorities.

Focus on high-potential prompts where you are absent or poorly placed. Also identify the strengths you should protect so you keep your lead.

A simple dashboard to help you visualize these results:

Prompt ChatGPT Perplexity Gemini Claude Average position

Best SEO tool Citation #3 Absent Mention Citation #1 2

SEO analysis software Mention Citation #2 Absent Mention 2

Semrush alternative Absent Citation #1 Citation #4 Mention 2.5

Method: how to find relevant prompts for your site

Building a useful prompt bank means combining several approaches. The goal is to match real user intent and the parts of the business that actually create value.

Turn your keywords into natural prompts

The first step is to start from your main SEO keywords and turn them into conversational prompts.

A keyword typed into Google, such as “CRM software”, then becomes “What is the best CRM software for an SME?” or “How do I choose CRM software?”.

That conversational form matches the way people query AI tools.

Transformation method:

Identify your priority keywords in your classic SEO strategy;
Turn them into natural questions (who, what, where, when, how, why);
Add contextual criteria: audience (SME, startup), place (Paris, France), budget, industry;
Cover different intents with varied formats.

Prompts such as “comparison”, “best X”, “where to buy”, “alternative to X”, “product for X”, or “service for X” often trigger a web search. With this kind of transactional query, AI tools try to recommend concrete solutions.

The long tail stays fully relevant in GEO. A very precise prompt such as “best CRM software for a real-estate agency with fewer than 10 agents” targets an ultra-qualified intent. Even at low volume, these prompts produce high conversion.

Define your prompts with dedicated tools

GEO solutions such as Peec AI, Qwairy, or Profound can give you lists of relevant prompts for your industry. They analyze frequent queries on LLMs and surface visibility opportunities. Those recommendations save a lot of time.

Sources and tools to identify your prompts:

Dedicated GEO tools: Peec AI, Qwairy, Profound for automatic suggestions;
Google itself: analyze People Also Ask (PAA) to find frequent questions;
Classic SEO tools: Semrush and Ahrefs let you filter keywords as questions;
Question tools: AlsoAsked, AnswerThePublic to explore what people ask;
The LLMs themselves: ask ChatGPT or Perplexity “what questions do users ask about [your topic]”.

Each tool brings a different angle and reveals complementary opportunities.

Analyze and select the prompts you found

Not every prompt you find is worth the same. This analysis phase lets you filter, classify, and prioritize the ones that truly deserve your attention.

The goal is to concentrate your effort on the most strategic prompts.

Prompts that trigger a web search

Not every prompt systematically triggers a web search by the AI models. Some produce a direct answer from the assistant’s knowledge base, with no external navigation. That distinction matters.

A prompt such as “Who is Albert Einstein?” usually does not trigger a web search. The information already lives in the model’s memory. By contrast, “What is the best CRM in 2025?” almost always triggers a web exploration to find recent reviews and up-to-date comparisons.

Questions to analyze for each prompt:

Does this prompt trigger a web search on ChatGPT, Perplexity, Gemini?
Which wordings trigger a web search: “best”, “comparison”, “review”?
Which themes, categories, or products systematically activate search?
Are there behavior differences between AI assistants?

There is little point hoping for a citation if the LLM does not run a web search. Put your resources on the prompts that actually trigger external exploration.

It still needs a nuance: if the absence of a web search blocks an outbound link (direct traffic), it does not mean there is no impact. You then distinguish two kinds of presence:

GEO visibility: the AI explores the web in real time to find fresh sources. This is where technical optimization of your content can win links and traffic.
AI branding: the AI cites you from “memory”, because your brand is part of its original training data (its knowledge cut-off).

Even without a blue link, being cited natively by a model such as Claude or GPT-4 as a sector reference is the highest stage of awareness. It strengthens your authority in the user’s eyes and indirectly shapes their later searches. A complete audit should therefore record these “spontaneous mentions”, which confirm the historical strength of your brand in the AI ecosystem.

Types of prompts

Classifying each prompt by its main intent helps you balance visibility across the full conversion funnel. That is how you refine your action plan and your optimization priorities.

Transactional prompts sit at the bottom of the funnel: “Buy SEO software”, “best competitive analysis tool”, or “cheaper alternative to X” target users who are ready to convert. These prompts generate less volume but a high conversion rate.

Informational prompts target the top and middle of the funnel. “How do I improve my SEO?”, “What is GEO?”, or “Complete digital marketing guide” build your authority and generate qualified traffic. They educate your audience and strengthen your expertise.

Analysis of your existing content:

Do you have content that answers each type of prompt?
Is that content relevant and up to date?
Does it clearly answer the search intent expressed?
Where does the answer sit in the content: beginning, middle, end?
Does the structure make extraction by LLMs easier?

This map reveals your gaps. A complete audit identifies those imbalances so you can adjust your content strategy.

Query fan-out: understand the terms behind your prompts

Query fan-out is a query-expansion method used by generative engines. From one prompt, artificial intelligence derives every sub-question and keyword needed to cover the topic thoroughly.

This technique shows how the AI assistant interprets your prompt. A user asks for “best CRM marketing campaigns”. The LLM automatically expands that into “best marketing automation CRM”, “CRM marketing comparison”, “CRM for B2B marketing campaigns”, “CRM with email”.

Questions to ask about your query fan-outs:

Are you present on the keywords identified in the query fan-outs of your prompts?
Do you have content that clearly answers those derived intents?
How does your brand rank on Google?
Do your pages cover every sub-topic the generative AI explores?

Analyzing query fan-outs increases your semantic coverage. The approach strengthens both your traditional SEO and your GEO. Content that answers the derived questions is more likely to be cited, because it covers the topic in depth.

Analyze your performance on the selected prompts

Measuring your performance on the identified prompts turns the audit into a concrete steering tool. This analysis shows your real position and guides your optimization plan to strengthen your online presence.

Measure your mention rate and citation rate

These simple, reliable KPIs tell you your real visibility in the answers generated by AI engines. They are the foundations of your GEO dashboard.

Mention rate: the percentage of selected prompts where your brand is named, with no associated link. For example, if ChatGPT mentions your company on 15 prompts out of the 50 tested, your rate is 30%. This metric evaluates your raw awareness in the eyes of LLMs.
Citation rate: the percentage of prompts where at least one URL from your domain appears as a source. If Perplexity cites your site on 10 prompts out of 50, your citation rate is 20%. This figure measures your authority and your ability to generate traffic.
Share of voice calculation: compares your visibility with that of your competitors on the same prompts. If 5 pages are cited on a prompt and you appear, your share of voice on that prompt is 20%.

Analyze these KPIs systematically by LLM. Your appearance rate on ChatGPT can differ sharply from Gemini or Claude. Those results steer your optimization work toward the platforms where you have the most room to grow.

Measure your positions in citations and mentions

Being visible is not enough. The position of your sources in generated answers directly affects click probability and the reliability users perceive. People place more trust in sources shown at the top of the answer.

Questions to analyze:

What is our average place in the citations?
Which competitors, domains, or URLs systematically rank at the top?
Does our rank vary by prompt type (informational vs transactional)?
Are there ranking differences between LLMs?

This competitive analysis identifies the players who dominate your sector. Study their content, their structure, their domain authority. Which elements explain their performance?

Beyond the citation: sentiment analysis

Being cited or mentioned by an AI is a technical win, but the business impact depends on the connotation of the generated message. Unlike a classic blue link on Google, the AI issues a judgment or a synthesis that directly shapes the user’s perception. A complete prompt audit should therefore include a sentiment analysis: is your brand presented in a favorable, neutral, or critical light?

If ChatGPT says your solution is “powerful but particularly complex to install”, the citation can paradoxically hurt conversion.

Sentiment evaluation criteria:

Overall tone: is the answer complimentary (strong recommendation), informative (neutral), or hedged with reservations (critical)?
Associated attributes: which adjectives systematically appear next to your name? (e.g. “expensive”, “innovative”, “robust”, “outdated”).
Direct comparison: in a comparative answer, does the AI place you as the “first choice” or as a “secondary alternative”?

Finding a negative connotation in AI answers is a major opportunity: it often means that the training sources (customer reviews, press articles, forums) carry a damaged image that you should correct with a new press-relations or e-reputation strategy.

Analyze your strengths and weaknesses

Identifying and understanding why you are present or absent on certain searches is an indispensable step if you want to improve your ranking.

Several dimensions are worth exploring. On which prompts are you visible or completely invisible? Are there particular topics, themes, or product categories where you excel, or where you simply do not exist?

Try to understand the origin of your performance:

Do the cited excerpts match the ones you expected?
Do they share common traits: structure, length, format?
Does your content answer the question asked in the prompt directly?

The gap with your competitors also sets your priorities. If you are absent on 80% of transactional prompts while three competitors dominate that ground, you have identified a major improvement axis. Invest in comparison content, richer product pages, and customer reviews.

Tracking your prompts and tools

Setting up regular tracking lets you steer your actions with data. That discipline turns a one-off audit into a continuous improvement process.

Visibility tracking is the base. Measure the evolution of your mention rates, citation rates, and share of voice on a defined cadence: weekly for very dynamic sectors, monthly for most companies.

Traffic tracking completes the analysis. How many visitors arrive from artificial intelligence? What is their conversion rate compared with other channels? These figures validate the business impact of your GEO strategy.

Available tracking tools:

Peec AI: automated prompt tracking and visibility analysis;
Profound: a complete dashboard with competitive benchmarks;
Qwairy: citation monitoring and alerts on ranking changes.

These solutions automate data collection and make reporting easier. They test your prompts regularly with the same protocol, which keeps your comparisons reliable over time.

Running a GEO prompt audit structures your approach to visibility on generative engines. Identify your key prompts, measure your performance, analyze your strengths and weaknesses: each step builds your new optimization strategy. Dedicated tools make it easier to track over time and turn this audit into a lever for improvement. In the era of artificial intelligence, this method becomes as essential as a traditional SEO (Search Engine Optimization) audit if you want to strengthen your positioning and credibility online.

Also discover how to run a GEO brand awareness audit to strengthen your credibility in the eyes of LLMs (and search engines!).

Sources used in this guide: