Signal · Measuring AI visibility
You can now measure how AI sees you. The number won’t fix what the answer says.
In spring 2026 tools such as Peec and Ahrefs started showing which AI crawlers reach your pages, how often your pages are cited and which hidden questions ChatGPT and Perplexity ask. That shows where you stand, but it doesn’t change what an AI answer says about your hours or services. And a US experiment posted in August 2026 found people clicked out to websites less when their searches went to Google’s AI Mode: being named, being right and winning a customer are separate things.
Published Updated
What changed
Three launches in one month made AI visibility countable
Peec, 27 March 2026
Citations page by page
A page for each URL: how often AI models retrieved and cited it, for which prompts, beside which brands, with the actual chats.
Ahrefs, 23 April 2026
Bot visits and hidden questions
Bot Analytics tracks crawler traffic in twelve categories, with an AI bots filter. Brand Radar shows the hidden follow-up questions (fan-out queries) ChatGPT and Perplexity generate before answering.
Peec, 27 April 2026
Crawl logs beside citations
Agent Analytics reads your server logs, through Cloudflare Workers or a CSV upload, and puts AI crawler visits next to retrieval and citation data.
Our read: measuring AI visibility is now real and competitive. None of these tools changes the facts an answer is built from.
What it means for you
A dashboard reads the answer. It doesn’t write it.
What a dashboard can show
- An AI crawler reached your page
- Your page was cited, how often, and for which prompts
- The hidden questions ChatGPT and Perplexity asked
- Where to look next
What only you can fix
- Whether the page it read was right
- Whether the hours, services, prices and policies it repeated are true
- Whether you answer those hidden questions well
- Correcting the source, which no measurement tool does for you
Peec itself says this data is directional, not cause and effect. In our reading, a spike in bot visits doesn’t prove a citation, and no log proves a recommendation.
Four ways a healthy dashboard hides a wrong answer
These are illustrative cases, not real businesses.
A restaurant sees a high citation rate and assumes it is winning, while the cited answer lists hours it changed two months ago.
A clinic watches AI bot visits climb and reads it as demand, when the crawled page describes a service it stopped offering.
An owner chases a hidden question with new content, while the pages AI already reads still carry the wrong address.
A business celebrates a spike in crawl traffic as proof it is being recommended, crossing the exact line the tool’s own documentation warns against.
In each case the dashboard is accurate and the conclusion is wrong.
Mention to customer
Five steps between being named and winning a customer
Each step can happen without the next. A single visibility score can’t tell you which one held and which one broke.
Named
The answer mentions your business.
What visibility reports show
Cited
The answer points to a source linked to you.
Referred
Someone leaves the answer for your site or page.
Committed
Someone calls, books or orders.
Became a customer
The booking or order is fulfilled and a real customer results.
Only your own systems show this
A Google Search experiment
In a US experiment, people assigned to AI Mode clicked out less
In March 2026 researchers at the University of Pennsylvania and Northeastern University ran a preregistered field experiment (its questions and analysis plan were published in advance) with US adults using Chrome and Google Search. For seven days each person had ordinary Google, Google with AI Overviews hidden, or every search sent to AI Mode. They counted clicks to websites, not customers. Percentage points are the plain gap between two percentages, not a percent of a percent.
Reported by the study’s authors, in a preprint not peer reviewed
18.8 percentage points
less external click-through when assigned to AI Mode
Compared with ordinary Google Search. It counts everyone assigned to AI Mode, where 94.7% of searches were redirected. Measured on searches by the 1,100 people who searched during the experiment.
Reported by the study’s authors, in a preprint not peer reviewed
0.92 sessions a day
fewer search sessions when assigned to AI Mode
Compared with ordinary Google Search, among the same 1,100 people.
Reported by the study’s authors, in a preprint not peer reviewed
8.8 percentage points
more external click-through where AI Overviews were hidden
Compared with unchanged Google, and counted only where hiding actually worked: across the experiment just 51.1% of AI Overviews were hidden, after Google changed its page code. It answers a different question, so it is not a mirror image of the AI Mode figure.
People assigned to AI Mode also rated search lower on trust, usefulness, satisfaction, control and relevance, among the 956 who completed the closing survey; hiding AI Overviews brought no detectable improvement.
What the study doesn’t tell you
It counted clicks to publishers and other websites. It didn’t study physical locations, local searches, Google Maps, menus, bookings, orders, purchases or revenue, and it didn’t follow anyone as far as becoming a customer.
It ran for seven days, so it can’t describe habits over months. Of 1,444 people who enrolled, 1,100 searched during the experiment and 956 completed the closing survey. The sample skewed younger, more educated and left-leaning, and AI Mode carried no ads during the study.
Researchers call the 18.8 figure an intent-to-treat estimate and the 8.8 figure a local average treatment effect. Both click figures are movements in percentage points, not percentage changes, measured on searches. The paper is a public preprint, not peer reviewed, and the raw browsing data are not released.
It also doesn’t show that correcting your business information earns a mention, a citation, a click, an order or revenue. Accurate hours and prices are worth having for their own sake.
What to do
Measure each step where it actually happens
Track crawler visits and citations separately
A bot visit proves a crawler arrived; a citation proves a page was used. One combined score hides where the chain breaks.
Read what the answer says about you
Check the hours, services, prices and policies it repeats, and correct any wrong one at the source you publish.
Use hidden questions as a to-do list
They show the follow-up questions ChatGPT and Perplexity ask themselves before answering. They tell you what to publish, not whether you already win.
Write down each observation in full
The AI system, the interface, the exact question, the language, the place, the date, the answer and the cited source.
Count visits, bookings and customers yourself
Read referrals in your own analytics and count calls, bookings, orders and completed customers where they are recorded.
A newer Briefing
A visible AI answer is not a customer
Being named in an AI answer is one event. Winning a customer is another. A US Google Search experiment measured clicks to publishers, not customers.
Read the full BriefingKeep reading
From measuring to correcting
Where this fits: the shift behind it, and the fix that happens at the source.
Questions about this research?
Write to Obenan Research about the sources, or about what they could mean for your business.
Sources
Where this comes from
Built from the vendors’ own changelog and blog, and a public preprint. Pages without a publication date show only the date we checked them. Obenan has no partnership, integration, endorsement, pilot or special-access relationship with Peec, Ahrefs, the study’s authors, their institutions, arXiv or Google.
- Randomized Google Search interface experiment (abstract and submission record)
arXiv Published Checked
The experiment’s design, preregistration and headline results. Submitted 18 August 2026.
- Randomized Google Search interface experiment (full paper)
arXiv Checked
Methods, denominators, the failed hiding condition, limitations and disclosures.
- Peec AI changelog: URL Details Page and Agent Analytics
Peec AI Checked
The URL Details Page (27 March 2026) and Agent Analytics with Crawl Insights and Crawlability (27 April 2026), and Peec’s statement that the data is directional, not causal. First checked 14 April 2026.
- New features, February to March 2026
Ahrefs Published Checked
Bot Analytics, the AI bots filter and Brand Radar’s fan-out queries.