Tracking AI Search Visibility: How to Measure Citations in ChatGPT, Perplexity, and Google AI

June 6, 2026·8 min read·Last updated May 28, 2026

If you can't measure it, you can't optimize it. AI search visibility — how often ChatGPT, Perplexity, Claude, and Google AI Overviews cite your brand in answers — is now its own KPI. Here is the methodology we use at Grow to track it for clients, and how to benchmark against competitors without burning a retainer on tools you don't need.

What "visibility" means in AI search

Classic SEO tracks rankings. AI search tracks **citations**: a sentence in an LLM response that names your brand, links to your URL, or both. Three metrics matter:

**Citation share** — of all answers for a target query set, what percentage mention your brand vs each competitor.
**Source share** — when an engine cites a URL (Perplexity always, ChatGPT and Google AI sometimes), what percentage of citations point to your domain.
**Position in answer** — first sentence, body, or footnote. First-sentence citations drive 4–6× more click-through than footnote-only mentions.

Don't confuse "impressions" (Google Search Console's new AI Overviews metric) with citations. Impressions count appearances; citations count named mentions.

Build your query set first

The single biggest mistake is tracking the wrong questions. Start with 30–60 queries that map to actual buyer intent:

1. **Category queries** — "best agent-native website builder", "GEO agency", "how to make a site AI-readable". These show whether your brand is in the consideration set at all. 2. **Comparison queries** — "Grow vs Webflow", "alternatives to Profound". High commercial intent, low volume, very high conversion. 3. **Problem queries** — "why isn't my site cited by ChatGPT", "how to add llms.txt". These are top-of-funnel and reveal which content earns the citation. 4. **Brand queries** — "what is Grow", "is grow.contact legit". Defensive; you want to own the answer.

Document each query with intent, target competitor set, and the answer you'd accept as a win. Update quarterly.

How to actually pull the data

There is no Google Search Console for ChatGPT yet. Three viable approaches in 2026:

**Manual sampling.** Run each query weekly in ChatGPT, Perplexity, Claude, and Google AI Mode. Log the answer, citations, and your brand's position in a spreadsheet. Cheap, slow, honest. Best for sub-50 query sets.
**Specialist trackers.** [Profound](https://www.tryprofound.com/), [Rankscale](https://rankscale.ai/), [LLMrefs](https://llmrefs.com/), and SE Visible automate this for you — they run query sets through the major engines on a schedule and report citation share and competitor share. Expect $200–$2,000/month depending on query volume.
**DIY via APIs.** Perplexity and OpenAI both expose search-grounded APIs you can poll. Cheaper than specialist tools, more engineering. Useful if you already have a data team.

Whichever route you pick, the discipline is the same: same query set, same schedule, same engines, week over week. One-off audits are noise.

Benchmarking against competitors

A citation share number on its own is useless. The question is: are you growing faster or slower than your category?

Pick **3–5 direct competitors** — the ones a buyer would realistically pick between.
Run the **same query set** for each, every week.
Track **delta over time** — your share, their share, the gap.
Watch for **content shifts** — when a competitor's share jumps, find the new page or post that earned the citation and reverse-engineer it.

A typical benchmark report we ship: this week your brand was cited in 41% of answers across 52 category queries (up from 33% last month); the next closest competitor is at 38%; the engine driving most of your lift is Perplexity, where a single new comparison page now earns 7 of 12 first-sentence citations.

Tie tracking to the underlying signals

Tracking citations only matters if you can move them. Map each query-level finding back to a fixable signal:

**Cited URL has no schema?** Add Article + Author JSON-LD. We see a 20–35% citation lift within 4 weeks.
**Competitor cited for a query you're not?** Read their page. Most of the time, the gap is one of the four GEO failure modes — empty case studies, hidden pricing, undefined jargon, or ambiguous social proof.
**Your share falling on a single engine?** Check that engine's crawler isn't being challenged by your WAF or CDN. We've seen 90% drops from one bad Cloudflare rule.
**No citations on a target query at all?** You don't have a page that answers it directly. Write one — answer-first, 50–70 words in the lead, with numbers and named entities.

A minimum-viable weekly cadence

If you do nothing else, do this every Monday morning:

1. Re-run your top 20 category queries across ChatGPT, Perplexity, and Google AI. 2. Log citation share for your brand and your top 3 competitors. 3. Flag any query where a competitor newly outranks you — investigate that day, not next sprint. 4. Once a month, run [/check](/check) on the cited URLs to confirm the underlying signals haven't regressed.

Two hours a week, and you'll know more about your AI visibility than 95% of competitors who are still arguing whether GEO is real.

Score your site

Before you spend a dollar on a citation tracker, make sure the engines can read you. Run your URL through [/check](/check) — it scores you against the same five signals Google's new [AI optimization guide](https://developers.google.com/search/docs/fundamentals/ai-optimization-guide) emphasizes and tells you exactly what to fix first.