How to measure AI visibility for an online store
AI visibility is not one number. It is three separate questions: can AI systems read your store, do assistants name you, and do people arrive and buy. Each question needs its own measurement and its own tool. A store can pass the first and fail the second, so measuring only one gives a false picture.
1. The three layers
| Layer | Question | What you measure | You control it? |
|---|---|---|---|
| Readable | Can AI systems reach and understand the store? | Crawler access, structured data, llms.txt, product feed | Yes, fully |
| Named | Do assistants mention or recommend you? | Mention rate, share of voice, citations | Partly |
| Visited | Do people arrive from AI and buy? | AI referral sessions, orders, crawler activity in logs | Partly |
The layers depend on each other in one direction only. A store AI cannot read will not be named. But a store AI can read is not automatically named – our own pre-registered study found that four technical signals did not separate often-named businesses from once-named ones.
2. Layer 1: can AI read the store?
This is the only layer you control completely, and the easiest to measure. Check from outside first, because that is what a crawler sees:
- Crawler access – does robots.txt allow the search crawlers, such as OAI-SearchBot, Claude-SearchBot and PerplexityBot? See GPTBot vs OAI-SearchBot and AI crawlers for Magento 2.
- Structured data – does each product page have JSON-LD Product with offers.availability, in the server HTML?
- Content map – does llms.txt exist, with a real summary?
- Feed – is an ACP product feed in place, if you want ChatGPT shopping results?
The free AEO scan checks this from outside in a few seconds. On Magento, bin/magento angeo:aeo:audit checks it from inside and can fail a CI build when the score drops.
3. Layer 2: do assistants name you?
This is what most merchants actually want to know – and the hardest to measure well. The measure is AI share of voice: out of the answers to a fixed set of buying questions, how many name you, and how many name your competitors.
Rules that make the number useful:
- Write the question set before you look at any results, and keep it fixed.
- Run each question more than once. Answers change between runs – in our study only 42% of businesses named once were named at all in the repeat run.
- Count mention, recommendation and citation separately.
- Keep answers with live web search apart from answers without it.
First-party sources exist, but they are limited. Bing Webmaster Tools reports citation counts under AI Performance. Google Search Console added generative AI reporting in June 2026, but it shows impressions, not citations. Neither covers ChatGPT, Claude or Perplexity answers.
On Magento, angeo/module-aeo-brand-visibility runs your question set across five assistants and scores competitors in the same run.
4. Layer 3: do people arrive and buy?
This connects AI visibility to revenue. It is also where data is most often wrong, because many AI referrals arrive with no referrer and are counted as Direct.
| Source | What it shows | Limit |
|---|---|---|
| GA4 “AI Assistant” channel | Sessions from recognised AI referrers, since May 2026 | Needs a referrer; not retroactive |
| GA4 custom channel group | Referrers outside Google’s list, plus history | You maintain the domain list |
UTM parameters (e.g. utm_source=chatgpt.com) | Some ChatGPT clicks that have no referrer | Only covers links that carry them |
| Server logs | AI crawler activity, a leading signal | Crawling is not traffic |
Step-by-step setup: How to track AI search traffic in Magento 2.
5. A monthly routine
- Readable: run the audit. Fix anything that went from pass to warning – stale files are the usual cause.
- Named: run the fixed question set twice. Record mention rate and share of voice against the last month.
- Visited: check AI Assistant and your custom channel in GA4, and the bot status codes in your logs.
- Write down what changed on the store that month. Without that note, you cannot link a change in the numbers to anything.
6. Common measurement mistakes
- Treating the audit score as visibility. It measures Layer 1 only.
- One run, one screenshot. AI answers vary; a single run is noise.
- Prompts that name your brand. They always find you and tell you nothing.
- Reading Direct traffic as “no AI traffic”. Part of it is AI with the referrer stripped.
- Claiming the cause. A number going up after a change does not prove the change caused it.
FAQ
What is the best way to measure AI visibility?
Measure three things separately: whether AI systems can read the store (technical audit), whether assistants name it (share of voice on a fixed question set), and whether people arrive and buy (AI referral traffic and orders).
Does a high AEO score mean ChatGPT will recommend my store?
No. The score measures whether the store is readable. Whether an assistant names it also depends on price, reviews and coverage on other sites.
Why does ChatGPT traffic show as Direct in GA4?
Many ChatGPT links, especially from the apps, arrive without a referrer, so GA4 cannot tell where they came from. UTM parameters and server logs help fill part of the gap.
How often should I measure?
Monthly is enough for most stores. Answers change, but not daily, and running more often mostly adds cost.
Can I measure AI visibility for free?
Yes. The web scan, GA4, Bing Webmaster Tools, Search Console and server logs are free. Running questions through AI assistants costs API fees on your own keys.
Written 22 September 2026. Dates for GA4, Bing and Search Console features are as reported on angeo.dev’s own guides; re-check vendor documentation before relying on them. Disclosure: we publish the open-source audit and brand-visibility modules referenced above.