How to measure AI search visibility in 2026
Ask ChatGPT, Gemini, Perplexity and Google's AI Overviews the same question and they'll cite the same domains in just 3.8% of cases, according to a 2026 Writesonic study (cited by Brand24). Read that again. "Are we visible in AI search?" isn't one question with one answer — it's four questions with four different answers.
So to measure AI search visibility you need three things: a tracked list of prompts your buyers actually type, a tool that checks several engines rather than one, and a small set of metrics watched as a trend instead of a single reading. Citation frequency, first-mention position and share of voice carry most of the weight. Referral traffic carries almost none, because the answers mostly resolve without a click.
If you haven't read our companion piece on AI search visibility and how to get cited, start there — this article doesn't re-cover getting cited. It's strictly about counting.
Why one engine's answer tells you almost nothing
The audience is split across platforms, and so is your visibility. ChatGPT now has 900 million weekly active users and Gemini passed 1 billion monthly users by August 2026 (HubSpot, 2026). Those aren't the same people asking the same things and getting the same sources.
With a 3.8% cross-platform citation overlap, a tool that checks one engine is measuring roughly one-quarter of your problem — and telling you it's the whole thing.
This is also why manual spot-checking fails. Answers vary session to session, so typing your category question into ChatGPT once a month produces false alarms: you vanish, you reappear, nothing actually changed. Automated, repeated sampling across engines — or noise.
Do that, and the picture usually looks lopsided — strong in ChatGPT, invisible in Perplexity, or the reverse. That's the real finding, and it's actionable. A single blended "AI visibility score" hides it.
What should you measure to measure AI search visibility?
Nine metrics do the job — but you don't need all nine in month one. The first three are enough.
| Metric | What it tells you | How to get it free or cheap |
|---|---|---|
| Citation frequency | Share of tracked prompts where your brand or domain appears | HubSpot AI Search Grader (free); AthenaHQ Essential (free) |
| First-mention position | Where you land in a multi-citation answer — being named first beats being named often | Any citation tracker; visible in free tiers |
| Share of voice | (Your mentions ÷ total category mentions) × 100 — 30 of 120 is 25% | Count manually from a tracker export, or Chatbeat/Otterly |
| Sentiment | Whether the AI frames you positively, neutrally or negatively | Read the actual answer text your tracker captures |
| AI referral traffic and assisted conversions | Directional signal that AI sent someone who converted | GA4 (free) — tag AI referrers, check assisted conversions |
| Time to first citation | How long new content takes to get picked up | Publish, then re-run the prompt weekly |
| AI Overview inclusion rate | Share of your target queries that trigger an AI Overview at all | Ahrefs' free Google AI Overviews tracker |
| Prompt coverage | Real conversation volume on a topic, versus keyword guesses | Paid tools only — skip at £0 |
| Content longevity | How long a placement keeps producing attention after publishing | Keep a dated log of your own tracker readings |
Two of these deserve more attention than they get.
Time to first citation gives you a publishing feedback loop. Typical content averages about six days to its first AI citation; press releases average eight hours (Notified research cited by CMI, 2026). Six days is a tolerable test cycle.
Content longevity is the one traditional analytics never captured. CMI's case for discoverability beyond blue links is that an AI placement keeps paying out long after the publish-week spike ends — so a 30-day performance window undervalues your best work.
Why your attribution will never be clean
Because the click is gone. Only 232 of every 1,000 U.S. Google searches now reach the open web — down from 360 two years earlier — and 68% end without a click at all (CMI, 2026). Ahrefs data puts a number on the damage: pages ranking #1 with an AI Overview present get 34.5% fewer clicks than #1 pages on matched queries without one (cited by Improvado, 2025).
No referrer, no session, no attribution. The influence happened anyway.
The honest approach is to stack several weak signals rather than chase one clean number:
- Server or crawler logs — which AI bots fetch what. A crawler fetch is not a human visit and not a recommendation. Don't report it as either.
- Self-reported attribution — one "how did you hear about us?" field on your signup or checkout form. Crude, free, and often the only place "ChatGPT" shows up at all.
- GA4 assisted conversions — imperfect and incomplete, but directional.
Pretending this adds up to precision is the biggest credibility risk in the whole discipline. Report it as a trend with error bars, not a KPI to two decimal places.
What do the tools cost, and which matter on a small budget?
Before the table, a disclosure: three of the six sources behind this article — Improvado, Brand24 and Profound — sell tools in this market, and Profound reviews itself among the 18 tools in its 2026 agency guide. Brand24's own case study, where share of voice climbed from 18–19% to 29% over four months (Brand24, 2026), is a vendor reporting on itself. Useful, not neutral.
All prices below were checked in October 2026 and will move. This market reprices constantly. Verify before you buy.
| Tool | Price (as of Oct 2026) | Best for |
|---|---|---|
| HubSpot AI Search Grader | Free; standalone AEO tier $50/mo | Your first ever reading |
| AthenaHQ | Free Essential tier; Starter $295/mo | Free tier plus GA4/Shopify attribution; tracks 8+ engines |
| Otterly.AI | From ~€29/mo | Small teams — fast setup, clear recommendations |
| AIClicks | From $59/mo | Cheapest paid multi-engine option |
| Chatbeat | $99/mo | Brand score, share of voice, sentiment, key sources |
| Semrush AI Visibility Toolkit | $99/mo per domain | Existing Semrush users (25 prompts, weekly refresh) |
| Similarweb | $99/mo | Teams already paying for Similarweb |
| GrackerAI | From $99/mo | Breadth — monitors 10 engines |
| Profound | $99–$399/mo | Daily tracking; prompt, answer and source analysis |
| Peec AI | $95–$495/mo | ChatGPT, Perplexity and Gemini monitoring |
| Scrunch AI | $250/mo core | Answer visibility plus technical crawler signals |
| Ahrefs Brand Radar | $199/mo per engine; $699/mo all engines | Agencies — plus a free AI Overviews tracker |
| Gauge | From $599/mo | Enterprise, API access included |
Most of that list is irrelevant to a solo creator. $99 to $699 a month to find out whether Perplexity mentions you isn't a defensible spend when you're running a business on your own — be honest about that before a vendor sells you a dashboard you'll check twice.
The £0 starting stack
Here's what to actually do if your budget is nothing:
- Write 30–50 prompts. Cover four types: brand ("is Doxiboo any good"), category ("best email tool for solo creators"), competitor ("X vs Y"), and problem-solution ("how do I automate cold outreach").
- Run them through HubSpot AI Search Grader and AthenaHQ's free Essential tier. Both cost nothing. Between them you'll see citation frequency and first-mention position across multiple engines.
- Add Ahrefs' free Google AI Overviews tracker for AI Overview inclusion on your target queries.
- Pull free impression data from Search Console and Bing Webmaster Tools, and tag AI referrers in GA4 for assisted conversions.
- Add one cheap tracker when free coverage frustrates you — Otterly.AI at roughly €29/mo is the sensible first paid step.
- Log the readings monthly in a spreadsheet and judge the trend, never a single month.
Don't wait for perfect infrastructure. Track share of voice and citation frequency in month one; add attribution later, when you've got something to attribute.
Where AI visibility measurement goes wrong
- Mistaking session variance for a lost placement — the single-check problem
- Trusting keyword-derived prompt lists, which miss how people actually phrase things
- Buying a monitoring tool that produces charts but never a decision
- Expecting full answer-engine coverage from a classic SEO platform
- Tracking one or two engines when the overlap is 3.8%
- Reporting visibility that never connects to revenue, signups or pipeline
- Judging content on a 30-day window and ignoring the long tail
- Counting an AI crawler fetch as a visit, or a mention as an endorsement
Frequently asked questions
What is the single most important AI visibility metric?
Share of voice, calculated as your mentions divided by total category mentions, times 100 (Brand24, 2026). It's comparative, so it survives the session-to-session variance that makes raw citation counts jumpy. Pair it with first-mention position — being named first in an answer beats being named often.
How many engines do I need to track?
Three or four minimum: ChatGPT, Gemini, Perplexity and Google AI Overviews. The 2026 Writesonic study found those four cited the same domains only 3.8% of the time (cited by Brand24), so single-engine tracking tells you very little about your overall position.
Can I measure AI search visibility for free?
Yes, enough to start. HubSpot's AI Search Grader and AthenaHQ's Essential tier are both free as of October 2026, Ahrefs offers a free Google AI Overviews tracker, and Search Console, Bing Webmaster Tools and GA4 cost nothing. That combination gets you citation frequency, AI Overview inclusion and directional conversion data without a subscription.
Why doesn't AI traffic show up properly in my analytics?
Because most AI answers end without a click — 68% of U.S. Google searches now end click-free, and only 232 of every 1,000 reach the open web (CMI, 2026). Your referrer data can only record the small fraction who click through, so it structurally undercounts AI influence.
How long before new content shows up in AI answers?
Around six days on average for typical content, with press releases averaging about eight hours (Notified research cited by CMI, 2026). Re-run your tracked prompts weekly after publishing rather than checking once and concluding nothing happened.
Final thoughts
Measurement here is young and genuinely messy. Nobody has a clean number, the tools disagree, and the engines disagree with each other 96% of the time. Don't wait for the industry to sort itself out — get a baseline this month with free tools, then watch the direction of travel.
And remember what you're measuring: rented attention. Every citation is an algorithm's decision that can reverse without notice, which is the argument we made for email marketing ROI. An email list is the one audience with nothing in front of it.
Doxiboo helps small businesses and solo creators turn those hard-won AI-search visitors into a list they own — AI-assisted sequences, no marketing team required. Start your free trial and stop renting your whole audience.
