Short answer: AI search visibility tracking works best as four measurements kept side by side: a fixed panel of 25–40 buyer prompts tested monthly in each AI engine (logging mentions, citations and accuracy), Google Search Console's Generative AI performance report for AI Overviews and AI Mode impressions, Bing Webmaster Tools' AI Performance report for Copilot citations, and a GA4 "AI assistants" channel group that sends AI referrals through to leads in your CRM. No tool gives a single reliable "AI rank", so report each number with its sample size and date.
Need this done for you? GrowthSparx offers a fixed-scope lead tracking and attribution setup: GA4, Google Ads, Meta, WhatsApp, calls and CRM wired into one dashboard, usually live in 2–3 weeks.
What should an AI search visibility tracking framework measure?
"Visibility" in AI search covers several different events, and mixing them is the most common reporting error we see. A brand named without a link sends no traffic; an AI Overview impression is not a click. Track each event separately.
| Event | Definition used in our sheets | Where it comes from | Why it matters |
|---|---|---|---|
| Mention | Brand or product named anywhere in the answer text | Prompt panel | Awareness and shortlist inclusion |
| Citation | A clickable link to a page on your domain in the answer or its sources list | Prompt panel; Bing AI Performance | The only route to a visit from that answer |
| Recommendation | Brand listed as an option to choose, not just referenced | Prompt panel | Closest proxy to "ranking" for commercial prompts |
| Accuracy | Facts about you (services, city, prices, licences) are correct and current | Prompt panel, manual review | Wrong prices or a closed branch cost enquiries |
| AI impression | Your link shown in an AI Overview or AI Mode response | Search Console | Google-scale exposure you cannot sample manually |
| AI referral | A session that arrives from an AI assistant | GA4 | Traffic you can tie to leads |
| AI-sourced lead | An enquiry whose first or last source is an AI assistant | CRM | The number the business actually cares about |
This post is about measurement only. If you have not yet checked that AI crawlers can reach your site, do that first with our AI search technical SEO audit; there is no point tracking visibility for pages that OAI-SearchBot or PerplexityBot cannot fetch.
How do you design a prompt panel for AI visibility tracking?
A prompt panel is a fixed list of buyer questions run on a schedule in each engine. Keep the wording fixed, or you cannot tell whether a change came from your work or from the prompt.
Size and mix
For a single-service business, 25–40 prompts is enough to see patterns without turning the monthly run into a two-day job. We split them roughly like this:
- Problem and learning prompts (25–30%): "why are my Google Ads leads not converting".
- Comparison prompts (25–30%): "SEO vs Google Ads for a dental clinic".
- Provider-choice prompts (30–35%): "best real estate lead generation agency in Noida".
- Brand prompts (10%): "is GrowthSparx good for B2B lead generation", "GrowthSparx pricing". These test accuracy, not discovery.
Write prompts the way Indians actually ask
Search Console query data is the best seed list, because it shows real phrasing. Add city and locality ("in Gurugram", "near Sector 62"), ₹ budgets ("under ₹50,000 a month"), platform names (99acres, Practo, IndiaMART, Shiksha) and one or two Hinglish versions for consumer categories ("Noida mein best physiotherapy clinic kaunsa hai"). AI assistants often answer these differently from the clean English version.
Control the conditions
- Use a logged-out or fresh session where the product allows, so chat history and memory do not personalise answers.
- Fix the location. ChatGPT uses an approximate location inferred from IP for local results, and its help article notes that a VPN can make that location wrong. Run local prompts from the city you are measuring, or name the city in the prompt.
- Record the engine and mode: ChatGPT with search on, Perplexity default, Google AI Mode, Copilot. Answers from a model without web search are a different thing; tag them separately.
- Run each prompt once per engine per month on a fixed date window (for example, the first three working days). Running a prompt five times and keeping the best result is how dashboards get inflated.
Example panel extract
Illustrative extract for a Noida-based agency selling real estate lead generation:
| ID | Prompt | Stage | Target page |
|---|---|---|---|
| P01 | How much does a real estate lead cost in India in 2026? | Learn | /blog/real-estate-lead-generation-cost-india.html |
| P02 | Should a builder use 99acres or Google Ads for leads? | Compare | /blog/property-portals-vs-google-meta-ads.html |
| P03 | Best real estate lead generation agency in Noida | Choose | /industries/real-estate.html |
| P04 | How to stop fake leads from Facebook property ads | Diagnose | /blog/reduce-fake-duplicate-real-estate-leads.html |
| P05 | GrowthSparx real estate results | Brand | /work/ileadguru-real-estate.html |
The target-page column shows whether the engine cited the page built for that question, another of your pages, or a competitor.
How do you score mentions, citations and accuracy?
Score each prompt-engine observation with simple, repeatable values so two people would record the same result.
| Field | Values | Rule |
|---|---|---|
| Mentioned | 0 / 1 | 1 if the brand name appears in the answer text |
| Recommended | 0 / 1 | 1 only if listed as an option to choose |
| Cited | 0 / 1 | 1 if any URL on your domain is linked |
| Right page cited | 0 / 1 | 1 if the cited URL equals the target page |
| Accuracy | 2 / 1 / 0 | 2 = correct; 1 = minor error or outdated detail; 0 = material error (wrong service, city, price or claim) |
| Competitors cited | Text | Domains cited instead of or alongside you |
From those fields, calculate a small set of metrics:
- Mention rate = observations with a mention ÷ observations run.
- Citation rate = observations citing your domain ÷ observations run.
- Right-page rate = right page cited ÷ observations with a citation.
- Accuracy rate = observations scored 2 ÷ observations with a mention.
- Citation share = your citations ÷ all citations to you and a fixed list of 3–5 named competitors.
Worked example (illustrative numbers): 30 prompts × 4 engines = 120 observations. You are mentioned in 18 (mention rate 15%) and cited in 12 (citation rate 10%). Eight of the 12 citations point to the intended page (right-page rate 67%). Of the 18 mentions, 14 are fully accurate (accuracy rate 78%), and four quote an old price. The action is obvious: fix the pricing page and the pages that repeat the old figure, before writing anything new.
Always print the denominator: "cited in 12 of 120 observations, 1–3 October 2026" can be checked next month; "cited in 10% of answers" cannot. With samples this small, treat a swing of two or three observations as noise.
Which first-party tools report AI search visibility?
Two search engines now publish AI-specific data for site owners. Neither covers ChatGPT or Perplexity, which is why the prompt panel and GA4 still matter.
Google Search Console: Generative AI performance report
Google's Generative AI performance report shows how many times links to your site were shown in AI Overviews and AI Mode. Key facts as checked on 11 October 2026:
- It reports impressions only; the help page does not list clicks, CTR or position for this report.
- Tabs: Pages, Countries, Dates and Devices, with a filter for text versus multimodal (Lens, Circle to Search) searches.
- Google says it was rolled out to all websites worldwide as of 31 August 2026. If you cannot see it, the property may not have enough AI impressions yet, or it is excluded from Search generative AI features in Settings.
- Search Labs experiments are excluded and the usual 1,000-row limit applies.
Export it monthly, filter Countries to India and compare each page's AI impressions with its Web impressions. Clicks from AI features are still counted inside the normal Performance report under the Web search type, as Google's AI features documentation explains, so you cannot isolate AI clicks in Search Console. For page-level fixes, see our Google AI Overviews optimisation guide.
Bing Webmaster Tools: AI Performance (public preview)
Microsoft launched the AI Performance report on 10 February 2026. It covers Microsoft Copilot, AI-generated summaries in Bing and "select partner integrations", and shows:
- Total citations of your site as a source in AI answers;
- Average cited pages per day;
- Grounding queries, the phrases the AI used to retrieve your content (a sample, not a full list);
- Page-level citation counts and a timeline.
It does not show placement within an answer. The grounding queries are the most useful part: they reveal the short retrieval query behind a long conversational prompt, which tells you what fact your page should state plainly.
Paid AI visibility trackers automate the prompt panel. Before trusting one, ask for its prompt list, engines, locations and run frequency; without the method, treat its score as a trend line.
How do you track AI referral traffic in GA4? (regex included)
GA4 has no built-in AI channel. Google's own custom channel group help page walks through creating one called "AI assistants" with a "matches regex" condition and placing it above Referrals. Its sample pattern is deliberately broad (any source ending in "ai" or containing "gpt", "gemini" or "copilot"), which can sweep in an ordinary referring site such as a news domain ending in .ai. We use an explicit list instead.
Step by step
- Admin → Data display → Channel groups → copy the default channel group and name it "Default + AI".
- Add a new channel named AI assistants. Condition: Source → matches regex → paste the pattern below.
- Drag "AI assistants" above Referral (and above Organic Search, so a stray Gemini referral is not counted as Google organic). GA4 puts traffic in the first channel it matches.
- Save. Google says custom channel groups apply to reports retroactively, so historical data is reclassified too.
- In Traffic acquisition, switch the primary dimension to your new channel group and add Landing page and Key events.
The regex:
^(.*\.)?(chatgpt\.com|chat\.openai\.com|perplexity\.ai|copilot\.microsoft\.com|copilot\.com|gemini\.google\.com|claude\.ai|chat\.deepseek\.com|meta\.ai|grok\.com)$
GA4's "matches regex" uses full matching by default, per Google's regex help page, so the pattern is anchored at both ends and the optional (.*\.)? covers subdomains such as www.perplexity.ai. Add domains as new assistants send you traffic; review the Referral channel's source list monthly for anything new.
What this does and does not capture
- ChatGPT: OpenAI's publisher FAQ says ChatGPT adds
utm_source=chatgpt.comto referral URLs, so those sessions show source "chatgpt.com" reliably. - Perplexity, Copilot, Claude, Gemini: appear as referrals when the browser passes a referrer. An app or browser that strips the referrer will send the visit to Direct.
- Google AI Overviews and AI Mode: clicks arrive as google / organic. Use Search Console for these, not GA4.
Sessions are only half the job. Store the source in a hidden form field and on the CRM record so a chatgpt.com lead keeps that label through to sale. Our UTM parameters guide covers the hidden-field setup and CRM revenue attribution covers carrying it to closed revenue.
What should an AI visibility tracking sheet contain?
One Google Sheet with four tabs is enough for most Indian SMEs and mid-size B2B firms. Copy these columns.
| Tab | Columns |
|---|---|
| 1. Panel | Prompt ID, exact prompt, stage, language, city, target URL, date added, status (core / exploratory / retired) |
| 2. Observations | Run date, prompt ID, engine and mode, location, logged in (Y/N), mentioned, recommended, cited, cited URL, right page, accuracy (0–2), error note, competitors cited, evidence link (screenshot) |
| 3. Platform data | Month, GSC AI impressions (India), top 10 pages by AI impressions, Bing total citations, average cited pages, top grounding queries, GA4 AI-assistant sessions, key events |
| 4. Outcomes and actions | Month, AI-sourced leads, qualified leads, revenue (if known), issue found, page changed, owner, date shipped, re-check date |
A pivot of tab 2 by engine and stage gives you the monthly scorecard. Tab 4 is the one leadership reads; if it is empty, the tracking is not yet connected to the business.
How often should you track AI search visibility?
| Cadence | Task | Time (typical SME) |
|---|---|---|
| Weekly | GA4 AI-assistant sessions and leads; check the CRM is recording the source | 15 minutes |
| Monthly | Run the core prompt panel; export Search Console Generative AI report and Bing AI Performance; update tabs 2–4; pick the top 3 fixes | 3–5 hours for 30 prompts × 4 engines |
| Quarterly | Refresh the panel (add or retire prompts); re-check crawler and firewall access; review competitor list; compare quarter on quarter | Half a day |
| After a major change | Re-run affected prompts 2–4 weeks after a pricing change, new service page or site migration | As needed |
Monthly suits the panel: weekly runs mostly measure answer variation, while quarterly is too slow to catch an outdated price being repeated.
How do you turn tracking data into fixes?
Every finding should map to a specific action on a specific URL.
| What the data shows | Likely cause | Fix |
|---|---|---|
| No citations in any engine, few GSC AI impressions | Access, indexing or thin pages | Run the technical audit; confirm pages are indexed with snippets allowed |
| Cited in Perplexity and Copilot, never in ChatGPT | OAI-SearchBot blocked at robots.txt or firewall | Check logs and WAF; see our ChatGPT search SEO guide |
| Mentioned but rarely cited | Brand known from third-party sites, but your own page lacks the specific answer | Add the direct answer, numbers and date to the target page |
| Wrong page cited | Two pages compete for one question | Merge or differentiate; link the weaker page to the stronger |
| Accuracy score 0 or 1 | Outdated facts on your site or on listings | Correct the source page, then Business Profile and directories |
| Competitor cited for data you do not publish | They have evidence you lack | Publish your own first-party figures with method and date |
| AI referrals rising, leads flat | Landing page does not convert this audience | Check the landing page's enquiry path and messaging |
For the content side of these fixes, our guide to content AI search engines cite explains what makes a page worth quoting. The wider strategy sits in our AEO and GEO guide, and if you would rather we set up the panel, channel group and reporting for you, see our AEO and GEO services.
Sources
- Google Search Console Help: Generative AI performance report: impressions-only metric, AI Overviews and AI Mode coverage, tabs, rollout to all sites by 31 August 2026, limits.
- Bing Webmaster Blog: Introducing AI Performance in Bing Webmaster Tools (10 February 2026): citation metrics, covered surfaces, sampled grounding queries.
- OpenAI Help: Publishers and developers FAQ: utm_source=chatgpt.com on ChatGPT referral links.
- Google Analytics Help: Custom channel groups: creation steps, AI assistants example, channel order, retroactive application, limits.
- Google Analytics Help: About regular expressions: full-match behaviour of GA4 regex.
- Google Search Central: AI features and your website: AI feature traffic counted in the Performance report under Web.
- OpenAI Help: ChatGPT search: approximate location from IP for local results; VPN caveat.