What Is AI Visibility Monitoring and How Does It Work?

Correction (6 October 2026): Our measurement provider withdrew two data sets used in this post. The 5 October 2026 scan was only partially measured: the provider ran out of credit during the scan; ChatGPT answered all 300 questions, while Gemini, Perplexity, Claude and Grok answered only 42–44. The Gemini and Google AI Overview zeros in the 15 September 2026 scan were also a measurement error. Those numbers in this post are invalid. The 15 September ChatGPT measurement (249 answers, 11.2%, ±3.9 points) stands. We will update this post when a complete new scan is available.

AI visibility monitoring is the work of measuring how often, where and through which source a brand is mentioned in the answers of generative engines such as ChatGPT, Gemini, Perplexity, Claude, Grok and Google AI Overview. Classic rank tracking counts the position of blue links. AI visibility monitoring measures presence inside synthesized answers that often contain no clickable link at all.

Short answer: Multi-engine monitoring rests on four things: a fixed query set, the engine set recorded with every scan, responses and margin of error per engine, and a list of sources showing which page got cited. In this post we explain how we set up this workflow for botfusions.com and which mistakes break the measurement.

1. Why classic rank tracking is not enough

There are two gaps:

• The dark funnel. If a user sees an AI recommendation and later comes to your site directly, analytics records the visit as "direct" or an untagged referral. The conversation that contained the recommendation appears in no dashboard.
• Probabilistic answers. Language models do not return a fixed ranking. The same question can surface different brands at different times. A single keyword position cannot show this volatility.

We showed with our own GA4 data why clicks alone fall short in our zero-click search post.

2. How AI engines choose sources

An engine that searches the web generally works in this order:

1. Query fan-out. The user's question is split into several sub-queries: comparison, pricing, reviews and so on.
2. Retrieval (RAG). Each sub-query hits a search index and relevant pages are fetched.
3. Passage selection and entity mapping. The engine picks relevant passages and recognizes entities such as brands, products and people.
4. Answer and citation. The model writes the answer and cites the sources it used, as links or by name.

This pipeline has two consequences for monitoring. First: the same model answers differently with web search on and off, so the report must state the condition under which it measured. Second: you also need to monitor which crawlers read your site. If crawlers such as GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot cannot reach your pages, content work will not show up in answers.

3. Multi-engine monitoring step by step

Step 1: Build the query set

Write the questions buyers actually ask: informational ("What is AI visibility monitoring?"), comparative ("X or Y?") and transactional ("Which GEO agency in Istanbul?"). Keep branded and unbranded questions apart. A branded question brings up the brand by itself; unbranded questions are the ones that measure visibility.

Step 2: Record the engine set with every scan

A measurement platform can change the engine set from scan to scan. In our September 15, 2026 scan ChatGPT, Gemini and Google AI Overview ran. In our October 5, 2026 scan ChatGPT, Gemini, Perplexity, Claude and Grok ran; Google AI Overview did not. We write the engines measured that day at the top of every report.

Step 3: Fix the cadence

We run a full scan once a month. In between, we run a small second measurement line where we document the model version and web search ourselves. Consistency matters more than frequency: same query set, same conditions.

Step 4: Read per engine

For each engine write three numbers: responses, mention rate, margin of error. A composite visibility score is a useful summary, but you should not make decisions on it alone. We explain how the score is calculated on our methodology page.

Step 5: Look at the cited pages

The mention rate answers "do we appear?". The citation list answers "with which page do we appear?". That list tells you directly which page to strengthen.

Step 6: Log the change, then measure again

Note the date and URL of every change you publish. In the next scan, count the effect as a change only on the same engines and only if it exceeds the margin of error.

4. Six mistakes that break the measurement

• Counting an unmeasured engine as 0%. An engine that did not run is "not measured". For Google AI Overview in our October 5, 2026 scan, that is all we can say.
• Comparing whole scans with different engine sets. Compare only engines that ran in both scans, one by one.
• Treating a small sample as a result. Gemini's 11.9% in the October 5, 2026 scan is 5 of 42 responses with a ±9.9 point margin. It is a signal, not a result.
• Branded queries inflating the rate. Most of the 25 queries in the citation sample of the October 5, 2026 scan were of the form "Botfusions or X". Do not trust the overall rate until you see the unbranded rate separately.
• Language mismatch. In the same citation sample, 13 of 25 citations went to our English /en/ pages, although most queries were in Turkish. Why the Turkish page was not chosen is a separate task.
• The wrong competitor set. Automatic competitor detection can bring in tools from another market. In the October 5, 2026 scan the top six by share of voice were US monitoring tools. Track the competitors of your own market separately.

5. A slice of our own monitoring

September 15, 2026 scan, ChatGPT: we were mentioned in 11.2% of 249 responses (±3.9 points). We withdrew the Gemini and Google AI Overview results of that scan and the October 5, 2026 table after our measurement provider reported an error.

We explain step by step how a scan becomes a report, how the comparison with the previous scan is done and why raw data is needed in our AI visibility report post.

6. Our measurement infrastructure

Our measurement infrastructure is a third-party platform. Scans give responses per engine, the margin of error and the cited URLs.

Alongside it we run a second measurement line. For every response it records the model version, whether web search was on, a timestamp and the full response text. On October 5, 2026 we collected 32 responses from 4 models on this line. The model that searches the web mentioned us in 3 of 7 unbranded queries. The three models without web search mentioned us in none. It is a small sample; it shows a direction, not a result.

7. A four-phase action plan

1. Audit access. Confirm that AI crawlers are allowed in robots.txt. Check that key content can be read without running JavaScript. Publish an llms.txt file.
2. Make content citable. Answer the question in the first paragraph. Tie every claim to a source or to your own measurement. The academic study on generative engine optimization measured that citing sources and adding statistics raise visibility; see our GEO implementation post for details.
3. Clarify the entity. Publish Organization, Article and FAQPage schema. Keep the organization description identical on every page. Point sameAs links to official profiles.
4. Monitor and iterate. Run the six-step workflow above once a month. Count a change only if it exceeds the margin of error.

Conclusion

AI visibility monitoring is replacing rank tracking in a search landscape where engines answer instead of linking. But a single score does not do the job. You need a workflow that shows which engines were measured, how many responses were collected, the margin of error and which page got cited. We monitor our own site with this workflow and share the results, good or bad, with the raw data.

Want to see where your brand stands in AI answers? Run your first measurement with Botfusions and get the engines measured, the margins of error and the raw data together.