AI visibility tracking is a measurement job: you sample AI answers on a schedule, log them in a fixed format, and compare each run with the last one.
Book a free SEO strategy session →
This spec covers the metric definitions, the prompt set, the run protocol, the log columns, the two first-party reports and the traps that quietly ruin the data.
If you want the engagement run for you, book a free SEO strategy session with Goldie Agency and bring this spec to the call.
Google is clear that nobody can promise a placement, so treat every number below as a reading, not a target someone can guarantee.
Lock the metric definitions before the first run
Most tracking sheets fail because two people count the same answer in two different ways.
Write these definitions at the top of the sheet and never change them mid-quarter:
- Mention: the brand string appears anywhere in the answer text, linked or unlinked.
- Citation: a URL on your domain is shown as a source, footnote or card. Peec AI points out that content can be cited without the brand being named, so log the two separately.
- Share of voice: your mentions divided by every brand mention across the same prompts and engines. Six of thirty is 20%.
- Sentiment: a three-point scale (positive, neutral, negative) plus a free-text note on the claim made about you.
- Prompt class: branded (contains your name) or unbranded (category, problem or comparison question).
- Engine: one value per surface. Google's AI Overviews and AI Mode are two values, not one.
Keep mentions and citations apart in every report. A brand that is named but never linked has a content problem, and a page that is linked but never named has a brand problem.
Build a 30-prompt set in four buckets
Thirty prompts is enough to see movement and small enough to run by hand.
- Category prompts (10): "best accountant for contractors in Bristol" style questions with no brand.
- Problem prompts (8): the pain a buyer types before they know a category exists.
- Comparison prompts (6): your brand against each named competitor.
- Reputation prompts (6): "is [brand] any good", "[brand] reviews", "[brand] alternatives".
Source the wording from sales call notes, support tickets and the queries tab in Search Console. Real phrasing beats invented phrasing every time.
Freeze the set for a quarter. Add new prompts in a separate tab so the trend line stays clean.
Book a free SEO strategy session →
🔥 Want a coach to sanity-check your prompt set and log? Inside the AI Profit Boardroom you learn AI SEO yourself through operator-level video lessons, an installable Agent OS for the grunt work, and four coaching calls every week to bring your tracking sheet to. → Join the AI Profit Boardroom
Run protocol: keep every sweep comparable
- Same account state: a clean session with no chat history or custom instructions, or the same logged-in test account every time.
- Same location: run from the market you sell in. A VPN hop changes local answers.
- Same window: run the whole sweep within one day.
- Repeat runs: fire each prompt three times per engine and record the majority result.
- Raw capture: paste the full answer text into a hidden column. You'll want it when a number looks odd.
Google explains on its AI features page that AI Overviews and AI Mode can use query fan-out, issuing several related searches across subtopics before writing a reply. Small wording changes therefore pull different sources, which is why the protocol matters.
The log schema that survives a quarter
One row equals one prompt, on one engine, in one run. These columns cover every report you'll need later:
| Column | Example value | Why it exists |
|---|---|---|
| run_date | 2026-11-16 | Trend lines and month-on-month deltas |
| prompt_id | CAT-04 | Stable key when wording is edited |
| class | unbranded | Splits reputation from discovery |
| engine | AI Mode | Per-surface reporting |
| mentioned | 1 | Mention rate |
| cited_url | /pricing-guide/ | Which page earns the link |
| competitors_named | BrandA; BrandC | Share of voice denominator |
| sentiment | neutral | Narrative tracking |
| note | called us 'small team' | Fact errors to fix at the source |
A pivot on engine and class gives you the headline table. A pivot on cited_url tells you which pages the engines actually trust.
📞 Got the sheet but not the time to read it? Book a free SEO strategy session and the Goldie Agency team will go through your first sweep, the competitors that keep getting named and the pages worth fixing first. → Get my free strategy session
Wire in Search Console's Generative AI report
Google's Generative AI performance report is the only first-party view of your AI Overviews and AI Mode exposure. Google says it rolled out to all websites worldwide as of 31 August 2026.
Book a free SEO strategy session →
- Open Search Console, then Performance, then the Generative AI report for Search.
- Set the date range to match your sweep window.
- Switch the table to the Pages dimension and export it.
- Filter search type to Web: text-based, then repeat for Web: multimodal if you sell visual products.
- Join the export to your log on cited_url.
Know the limits. The report counts impressions, meaning times a link to your site was shown. The help page lists Pages, Countries, Dates and Devices as dimensions, and no query or click columns.
Chart totals aggregate by property, so two of your links in one AI answer count once. Search Labs experiments are excluded, and the newest days can be preliminary.
Wire in Bing AI Performance
Bing's AI Performance report covers Microsoft Copilot, AI summaries in Bing and select partner integrations.
- Total Citations and Average Cited Pages give the volume line.
- Grounding queries are a sample of the phrases the system used to retrieve your content. Feed them back into your prompt set.
- Page-level citation activity slots straight into the cited_url join.
- Citation Share, added in June 2026, is your citations as a percentage of all citations for the same grounding query.
Bing states the counts don't indicate ranking, placement or authority, and that Citation Share does not expose competitor domains. Report them as observations, never as positions.
Variance: how many repeats before you act
Expect the same prompt to wobble between runs. Google says AI Mode and AI Overviews may use different models and techniques, so even two Google surfaces won't agree.
Bing lists evolving AI models, competing content, freshness and shifting demand as reasons citation activity moves.
A workable rule for small sets: act on a change only when it holds across two consecutive sweeps and at least two of three repeat runs.
Report rates with the sample size beside them. "Mentioned in 7 of 30 unbranded prompts on ChatGPT" is more useful than "23% visibility".
Crawler checks when an engine never cites you
A zero across every prompt on one engine is usually an access problem, not a content problem. Check robots.txt and the CDN bot rules for the search crawlers:
Book a free SEO strategy session →
User-agent: OAI-SearchBot
Allow: /
User-agent: PerplexityBot
Allow: /
User-agent: Claude-SearchBot
Allow: /
OpenAI says sites opted out of OAI-SearchBot won't be shown in ChatGPT search answers and that robots.txt changes take about 24 hours. Perplexity says PerplexityBot surfaces and links sites. Anthropic says disabling Claude-SearchBot may reduce visibility in search results.
Training crawlers such as GPTBot and ClaudeBot are separate rules. Blocking them doesn't remove you from search answers.
Tools that run the sweep for you
Past 30 prompts or three engines, manual runs get expensive in hours. These products automate the sampling, each described from its own site:
- Ahrefs Brand Radar: stores responses from six AI surfaces, Google's two among them, and benchmarks share of voice and sentiment against rivals.
- Semrush AI Visibility Toolkit: a Prompt Tracking module for your chosen prompts on AI Mode and ChatGPT, next to Brand Performance reports.
- Otterly.AI: logs mentions and citations on seven engines and lets you tag prompts Branded or Non-Branded, which maps onto the class column.
- Peec AI: segments by model, country and custom tag, and schedules runs daily, weekly or monthly.
- SE Ranking AI Visibility Tracker: checks whether a domain is mentioned or linked in AI Overviews, AI Mode, ChatGPT and Gemini, beside competitors.
Ranking them is a different job. Our list of the best AI visibility tools does that. Whatever you buy, Google's generative AI guide states that outside vendors have no window into its internal ranking or AI systems.
Where AI agents speed this up
- Prompt drafting: an agent turns a folder of call notes into a first 30-prompt draft for you to edit.
- Answer scoring: paste raw answers in and let the agent fill mentioned, cited_url, competitors_named and sentiment against your written definitions.
- Delta reports: the agent compares this month's pivot with last month's and lists the moves that pass the two-sweep rule.
- Draft-only rule: the agent never edits the live site; it hands you a list.
Spot-check one row in ten by hand. Agents misread sarcasm and comparison tables just like people do.
Reading the first delta report
The second sweep is where tracking starts to earn its keep, because you finally have something to compare.
Book a free SEO strategy session →
Start with unbranded mention rate per engine. A drop on one engine while the others hold steady points to something engine-specific: a crawler rule, a lost source page or a model change.
Next, sort the cited_url pivot by change. Pages that gained citations show which format the engines prefer for your topics, so copy that structure onto the weaker pages.
Then read the note column. Every wrong fact has a source somewhere: an outdated directory, an old review site or your own stale About page.
Finish with share of voice by competitor. If one rival climbs on every engine in the same month, look at what they published or where they were featured.
Gotchas that corrupt the data
- Editing a prompt's wording without giving it a new prompt_id.
- Counting a citation to a third-party review of you as your own citation.
- Running half the sweep on Monday and half on Friday.
- Treating one Google AI Overview that didn't trigger as a loss; Google says they often don't show at all.
- Mixing branded and unbranded prompts into one visibility number.
- Calling a Citation Share figure a rank.
When a result looks wrong, read the raw answer column before you change the site.
📞 Want the whole tracking stack set up for you? Book a free SEO strategy session to map out the prompt buckets, the log and the report joins with the Goldie Agency team, and leave with the fixes your data points to. → Book my free strategy session
My verdict: the minimum stack worth running
Run a frozen 30-prompt log across ChatGPT, AI Mode, AI Overviews, Gemini and Perplexity every month, join it to the Search Console and Bing exports, and only act on moves that hold for two sweeps.
Add a paid tracker when the manual run passes two hours, or book a free SEO strategy session and hand the sweep to Goldie Agency.
If you'd rather learn the method in depth, the AI Profit Boardroom is where to learn AI SEO yourself.
📘 Free resources for the next sweep Grab the free 200+ AI SEO Prompts to seed your prompt buckets, and the free link building book for the mentions work that moves share of voice. When a sweep shows a deeper problem, run the full AI visibility audit.
Book a free SEO strategy session →
Also On Our Network
- 🌐 A plain-English monthly routine for owners checking if AI recommends them
- 🌐 Six AI tracking methods compared by what each one can actually see
- 🌐 Three Agent OS prompts that build, score and compare an AI tracking log
- 🌐 Why each AI engine needs its own column in your tracking log
- 🌐 The complete AI visibility tracking guide on AI Profit Boardroom
FAQ: AI Visibility Tracking
What is the difference between a mention and a citation?
A mention is your brand name in the answer text. A citation is a link to your page shown as a source. Log them in separate columns because they point to different fixes.
How many prompts should an operator start with?
Thirty is a practical start: ten category, eight problem, six comparison and six reputation prompts. Freeze the set for a quarter so the trend stays readable.
Does the Search Console Generative AI report show queries?
Google's help page lists Pages, Countries, Dates and Devices as its dimensions and describes impressions, not queries or clicks. Use your own prompt log for query-level detail.
Is Bing Citation Share a ranking?
No. Bing calls it an observational metric: your share of all citations for one grounding query. It does not expose competitor domains or traffic share.
How often should the sweep run?
Monthly for a manual log. Tools can run daily or weekly, which helps once you track dozens of prompts or several markets.
Can a tracking tool read Google's own AI data?
No. Google rules that out: outside vendors have no access to its ranking or AI internals. Search Console is the only place Google shares data about your site.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link building agency, Goldie Agency, with a team of 70+ SEO experts.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote "Link Building Mastery".
- My Udemy courses have taught over 29,000 students.
→ Get my best AI training inside the AI Profit Boardroom
Freeze the prompts, keep the protocol tight, join the first-party exports, and line up a free strategy session the day you'd rather have Goldie Agency operate your AI visibility tracking.











