SEO Automation
AI Search Visibility Tracking From Real Prompts and Saved Answers
I run a list of the questions your buyers actually ask through the main AI answer surfaces, in a real browser, and save what each one says. For every prompt you get a screenshot, the answer text and a clear record of whether your business was named or cited as a source.
- years in SEO
- 6+
- buyer prompts tracked in AI answers
- 160+
- person doing the work
- 1
What AI search visibility tracking tells you
AI search visibility tracking, as I run it, means putting a fixed list of real buyer questions to the AI answer surfaces people use, saving every answer, and recording whether your business was named or cited. It answers a simple question that analytics cannot: when a buyer asks an AI tool for help, does your business come up?
More buying journeys now start with a question typed into an AI tool rather than a list of links. Those answers rarely send a measurable click, and they do not come with positions you can track in a spreadsheet. What you can do is ask the questions yourself, systematically, and keep the evidence. Done by hand, that means hours of copying prompts into four different tools and taking screenshots. The pipeline does the asking and saving; I do the reading.
It is one of the SEO automation pipelines I built and maintain, written mostly in Node with Playwright driving the browser. The output is evidence you can audit yourself, not a single visibility score.
Key takeaways
- Prompts are written from real buyer questions, not invented to flatter the brand.
- Each prompt runs through Google's AI results, Gemini, ChatGPT and Claude in a real browser, the way a person would see them.
- Every answer is saved as a screenshot and as text, with a record of whether your brand was named, cited as a source, both or neither.
- Results come as a browsable gallery and a CSV you can filter, share and compare with later runs.
- AI answers change from one session to the next, so this is a dated snapshot of what buyers could see, not a rank tracker.
What's included in my AI search visibility tracking
A tracking run has a few distinct parts, from writing the prompts to the files you keep. Each tab covers one of them.
A prompt list built from the questions your buyers really ask
The run is only as useful as the prompts in it. I build the list with you from the questions prospects ask on calls, in emails and in search: service plus location, comparisons between options, cost questions, and requests for a recommendation. Prompts that name your brand are kept separate, because they test something different from the open questions where a buyer has not chosen yet.
- Prompts drawn from real sales and support questions
- Service, location, comparison and cost questions covered
- Branded and unbranded prompts kept apart
- List agreed with you before the first run
Google AI results, Gemini, ChatGPT and Claude covered in one run
Buyers do not all use the same tool. Each prompt is run through Google's AI-generated results, Gemini, ChatGPT and Claude, so you can see where your business shows up and where it does not. The surfaces often disagree with each other, and that disagreement is useful information in its own right.
- Google's AI-generated search results
- Gemini answers
- ChatGPT answers
- Claude answers
Prompts run in a real browser, the way a buyer sees answers
Answers are collected by driving a real browser through each surface rather than calling a developer API. API responses can differ from what the consumer product shows, and the consumer product is what your buyers use. The browser approach is slower and fussier, and it records what a person would actually have seen.
- Consumer interfaces used, not developer endpoints
- Answers captured as rendered on screen
- Source chips and links recorded as shown
- Same approach on every surface
A saved screenshot and full answer text for every prompt
For every prompt on every surface, the run saves a screenshot of the answer and the answer text itself. The screenshot is the proof, useful for a board report or a client presentation. The text is what makes the results searchable, so you can find every answer that mentions a competitor or a particular claim about your business.
- Screenshot saved per prompt and surface
- Full answer text stored alongside
- Searchable across the whole run
- Evidence you can share without rerunning anything
Brand named, cited as a source, both or neither recorded
Being mentioned in an answer and being used as a source are different outcomes. The run records whether your brand is named in the answer text, whether your site is cited as a source, or both. A brand that is cited but never named is being used for facts without getting the recommendation, and that calls for a different fix than a brand that is absent altogether.
- Named in the answer recorded per prompt
- Cited as a source recorded per prompt
- Both outcomes kept as separate fields
- Screenshot available to confirm each verdict
A browsable gallery and a CSV of results you can work with
Results arrive in two forms. A gallery page lets you scroll through every screenshot by prompt and surface, which is the fastest way to get a feel for how AI tools describe your market. A CSV holds the same results as rows, so you can filter by surface, by prompt type or by outcome, and compare against a later run.
- Gallery of every screenshot, grouped for browsing
- CSV with one row per prompt and surface
- Filterable by surface, prompt and outcome
- Files you keep and can share internally
Paced, resumable runs that cope with platform limits
AI platforms rate-limit and challenge automated use, and some stop answering after a number of questions in a short period. Runs are paced to stay within what each surface tolerates, and they resume where they stopped, so an interruption costs time rather than the whole run. Failed prompts are retried rather than silently dropped.
- Requests spaced out per platform
- Runs pick up where they stopped
- Failed or empty answers retried
- Gaps reported rather than hidden
Experience
AI Search Visibility Tracking from someone who has done it many times
You work with me from the first call to the last report. No account managers, no hand-offs to a junior team.
Book a free call- years in SEO
- 6+
- client websites worked on
- 60+
- buyer prompts tracked in AI answers
- 160+
- documented case studies
- 35
Why AI answers need their own kind of tracking
Traditional rank tracking assumes a stable list of results with positions. AI answers break most of that assumption.
No fixed positions
An answer is a paragraph, a list or a comparison, not ten ranked links. The useful questions are whether you appear at all, and in what role.
Answers vary between sessions
The same prompt can produce a different answer an hour later. A single check proves little; a set of prompts across surfaces shows a pattern.
Mentions and sources differ
An AI tool can quote your site as a source without recommending you, or recommend you without linking. Both are worth knowing, and they need different work.
Each surface behaves differently
Google's AI results, Gemini, ChatGPT and Claude draw on sources in different ways. Visibility in one says little about the others.
How a tracking run is set up and delivered
A first run usually takes one to two weeks from agreeing the prompts, depending on the size of the list and how heavily the platforms throttle that week.
- 1
Gather buyer questions
I start from the questions your prospects actually ask. I draft prompts from them and from the searches your pages target, and you add any you hear on sales calls.
- 2
Agree the prompt list and brand rules
The final list is fixed before the run, along with the brand names, domain and spellings that count as a match, so results are consistent and comparable later.
- 3
Run each surface
The pipeline works through the list on each surface in a real browser, pacing requests, saving screenshots and answers, and resuming after any interruption.
- 4
Record named and cited outcomes
Each answer is checked for your brand in the text and for your site among the sources. I review uncertain cases against the screenshot.
- 5
Build the gallery and CSV
Results are assembled into a gallery for browsing and a CSV for filtering, with gaps or failed prompts listed openly.
- 6
Walk through the findings
I go through the patterns with you: where you appear, where competitors appear instead, which sources the tools lean on, and what that suggests for the next piece of work.
Free review
Get a free AI Search Visibility Tracking review
Send me your site. Before the call I look at your rankings and pages, then tell you what I would fix first and what it is likely worth.
Reading the outcome of each prompt
Every prompt on every surface ends in one of four outcomes. Each points to a different next step.
| Outcome | What it looks like in the answer | What it usually suggests |
|---|---|---|
| Named and cited | Your business is recommended and your site appears as a source | Strong position for that question; protect the page being cited |
| Named, not cited | Your business is mentioned, but sources are other sites | Others describe you; your own pages may not answer the question clearly |
| Cited, not named | Your page is used as a source, but the answer recommends others | Your content informs the answer without making the case for you |
| Neither | You do not appear in the text or the sources | A gap; the screenshot shows who the tool relied on instead |
These are readings of a snapshot. A single prompt in a single session is weak evidence; patterns across many prompts and surfaces are what matter.
A keyword rank tracker compared with prompt-based answer snapshots
Some tools now sell AI visibility as a score that rises and falls. A score is easy to chart, and it hides what was actually said.
Rank-style AI score
- One number that blends many answers together
- Prompts chosen by the vendor, not your buyers
- Little or no view of the answer itself
- Implies a stability AI answers do not have
- Hard to check or explain to a board
My prompt snapshots
- Every answer saved as a screenshot and as text
- Prompts built from your real buyer questions
- Named and cited recorded as separate outcomes
- Presented as a dated snapshot, honestly labelled
- Raw files you can audit and share
Measuring versus earning
Tracking shows where you stand; it does not change it
This pipeline measures visibility. Earning it is a different job: making your pages the clearest answer to the questions buyers ask, and getting described accurately on the sites AI tools draw from. That work is AI search optimisation, and a tracking run is a sensible first step before it and a way to check progress after.
Limits to keep in mind when reading the results
AI answers change. They vary between sessions, can depend on location and account state, and shift as the underlying models and sources are updated. A run records what the surfaces said on the day for the prompts asked. Rerunning the same list later shows how the picture moved between two snapshots, which is useful, but it is not the same as a daily position chart and I will not present it as one.
Platforms also limit automated use. Some throttle or challenge repeated questions, which is why runs are paced and resumable and why very large prompt lists take longer. If a surface blocks a set of prompts, those gaps are listed in the results rather than filled with guesses.
How tracking connects to the rest of your search work
Most of what earns a mention in AI answers is ordinary good SEO done thoroughly: pages that answer questions clearly, accurate information about your business across the web, and sources worth citing. That is why tracking often leads into AI search optimisation, content strategy for the questions where you are missing, or off-page SEO when the tools rely on third-party sites that describe your market without you.
If the results show competitors appearing where you do not, competitor analysis automation helps explain what their sites have that yours lacks. For the full set of pipelines, see SEO automation services, or get in touch with a few questions your buyers ask.
Growth from real projects
Learn more about ai search visibility tracking
AI Search Visibility Tracking FAQs
Which AI tools does your AI search visibility tracking cover?
Google's AI-generated search results, Gemini, ChatGPT and Claude. Each prompt runs through every surface in a real browser, and the results are kept separate per surface, since the tools often give different answers to the same question and draw on different sources.
Is this a rank tracker for AI search?
No. AI answers have no fixed positions and change between sessions, so I present the results as a dated snapshot of what each surface said for each prompt. Rerunning the same prompts later shows how things moved between snapshots, but it should not be read as a daily ranking.
What is the difference between being named and being cited?
Named means your business appears in the answer text, usually as a recommendation or example. Cited means your website is listed as a source the answer drew on. You can have one without the other, and each points to different work, so the results record them separately for every prompt.
How many prompts should a run include?
Enough to cover the main ways buyers ask about what you sell: your core services, locations, comparisons and cost questions, plus a few branded prompts. For most businesses that is a few dozen to a few hundred. Larger lists take longer because platforms limit how many questions they will answer in a short period.
Why use a real browser instead of the AI companies' APIs?
Developer APIs can return different answers from the consumer products, and they often leave out the sources and layout people see. Buyers use the consumer products, so that is what the run records. The trade-off is speed: browser runs need pacing and can be interrupted, which is why they are built to resume.
Can you track competitors as well?
Every screenshot and saved answer shows which businesses the tool named and which sites it cited, so competitors are visible throughout the results. Searching the answer text for a competitor's name shows where they appear and you do not, which is often the most useful part of the walkthrough.
Start AI search visibility tracking with your own buyer questions
Send me five to ten questions your prospects ask. I will run them through the AI surfaces and show you the screenshots, so you can see where you stand before deciding on a full prompt list.