How Do You Know If AI Search Is Sending You Buyers?

To measure AI search visibility, track a minimum of 50 prompts per site across ChatGPT, Google AI Overviews and Google AI Mode, and monitor two core metrics: mention rate (how often you appear at all) and competitive win rate (how often you’re mentioned without a competitor in the same answer). Pair this with a simple habit: ask “where did you find us?” on every sales call, since AI-driven enquiries rarely show up cleanly in analytics.

Why Click-Based Tracking Undersells AI Search

The click collapse.

Position 1 CTR when an AI Overview appears (Ahrefs, 300k keywords)

7.3%

1.6%

Dec 2023

Dec 2025

-78%

position 1 click-through under an AI Overview, up from -34.5% a year earlier (Ahrefs, Dec 2025)

-51%

position 2 falls almost as hard, and position 3 drops 46% (Ahrefs)

~93%

zero-click rate inside Google AI Mode (Semrush)

If you’re judging AI search purely on clicks, the numbers look bad. Position 1 click-through when an AI Overview appears has dropped 78% since December 2023. Zero-click rate inside Google’s AI Mode sits around 93%. Most marketing dashboards built around clicks and sessions will show a lot of nothing.

But citations count even without a click. Brands cited in an AI Overview see 35% more organic clicks elsewhere. Brands named directly in ChatGPT answers get 2.5 times more direct visits. A prospect who sees your name in an AI answer, closes the tab, and searches your brand name directly a day later won’t show up as an AI referral anywhere in your analytics, but the AI mention is still what put you there.

This is the core problem: most construction, manufacturing and property marketers genuinely don’t know whether ChatGPT or AI Overviews mention their business, because nothing in a standard analytics setup is built to tell them.

What to Actually Track

Start with a simple baseline: a minimum of 50 prompts per site, covering (at minimum) ChatGPT, Google AI Overviews and Google AI Mode. If budget allows, extend this across other assistants too.

Track the prompts.

Track your prompts across all the LLMs if budget allows. At minimum, track ChatGPT, Google AI Overviews and Google AI Mode for a view of performance.

65

Prompts

Within selected range

11,193

Responses

Within selected range

46%

Mention Rate

5,181 of 11,193 responses

17.4%

Competitive Win Rate

Brand mentioned without competitor mention

Brand Mentions Over Time (Past 30 Days)

Mention Rate (%)
Mention Count
Active Prompts
DNA Kids

50%
37%
25%
12%
0%
Jun 06
Jun 20
Jul 05

Top Performing Prompt

“My child wants a science-themed party. Who are the best entertainers to provide this?”

Brand rate 94.4% (177 responses)

Top Performing Assistant

Perplexity

Brand rate 55.7% (1,954 responses)

Best Performing Tag

Theme Or Type Based

Brand rate 71.8% (2,030 responses)

Two metrics matter more than the rest when you measure AI search visibility:

  • Mention rate: how often your brand shows up at all across tracked prompts
  • Competitive win rate: how often you’re mentioned without a competitor appearing in the same answer

 

A high mention rate alongside a low competitive win rate usually means you’re getting lumped in with the pack rather than recommended outright.

It’s also worth reviewing domain citations: which third-party sites are actually being cited when your sector gets discussed. This tends to surface competitors, review platforms, and industry directories you hadn’t been watching closely.

Your website is only half the story

LLMs do not just rank pages. They form an opinion of a brand from everything they have read about it, across the whole web.

When an AI recommends a company, it leans heavily on review sites, press coverage, forums and ‘best of’ lists. In studies of AI answers, most citations point to third-party sites, not the brand’s own website.

This is not new. Off-site authority was always the hard part of SEO.

AI often trusts what others say about us more than what we say about ourselves.

Where AI learns about us

AI’s view of
our brand

Our website

Reviews

Forums &
social

Directories & maps

News & PR

Navy = places we do not control directly

Prompt Research: Finding Out What People Are Actually Asking

Before you can track the right prompts, you need to know what people are asking AI in the first place. Google Search Console has a filter for this. Set a custom regex filter on your top queries using something like what|where|how|why|when|should, and you’ll pull out every question-style query your site already appears for.

Google Search Console: filter by query.

Search results > Filter by Query

Query

Filter
Compare
By keyword
Custom (regex)
Matches regex
How to use regex
Enter regular expression (regex)
what|where|how|why|when|should
Branded queries ?
Non-branded queries ?
Cancel
Apply

Run that export through an LLM to group the results by intent, and prioritise the commercially led ones (a query like “how do I compare structural glazing suppliers” matters more to your pipeline than “how does structural glazing work”). This is directional work rather than a fixed formula: the goal is understanding the shape of demand, not chasing every possible phrase.

Turning Tracking into Action: The Coverage Report

Once you know what people are asking and how often you’re mentioned, the useful next step is a simple coverage report: one row per prompt, showing the best matching page on your site, and a verdict on what to do about it.

Create a coverage report

One row per prompt. Instantly shows what to fix and what to create.

Prompt we want to appear for

“How do I choose a supplier for X?”

Best matching page

/guides/choosing-a-supplier

Verdict

Strong: polish it

Prompt we want to appear for

“What does service Y cost?”

Best matching page

/services/overview

Verdict

Partial: rework page

Prompt we want to appear for

“Is X better than Y for small firms?”

Best matching page

(nothing close)

Verdict

Gap: create content

A prompt with a strong matching page just needs polish. A partial match means an existing page needs reworking to actually answer the question. No match at all is a genuine content gap, and usually the most valuable one to close, since it’s ground a competitor hasn’t covered either.

Five steps to optimise for AI search.

1

Discoverability

Can AI find and crawl your websites?

2

Prompt Research

Do you know what users may be prompting in AI Search.

3

Data Collection & Analysis

Where do you get cited and where else could your brand be positioned with supporting content?

4

Content Analysis

Assess your own content: does it answer these questions?

5

Continued Monitoring

Do not stop measuring. Just like traditional Google organic rankings, you need to know when things change.

The Low-Tech Check That Catches What Dashboards Miss

None of the above catches every AI-driven enquiry. A simple habit fills the gap: make sure sales teams, yours and your clients’, are asking “where did you find us?” on every call. It’s unscientific, but it consistently surfaces AI-driven leads that never show up as a trackable referral in any platform.

Don’t Stop Once You’ve Set This Up

Just like traditional rankings, AI visibility moves. One ChatGPT model change halved AI referral traffic for some sites overnight in November 2025. Treat this as ongoing measurement, not a one-off audit.

For a broader look at how AI search fits into your overall reporting, see our complete guide to construction analytics and our piece on choosing the right analytics platform. If you’re already unsure whether your current SEO reporting tells you anything useful, this one’s worth a read too. And for the wider context behind all of this, see our GEO vs SEO explainer and AI shakeup piece.

If you want help setting up proper AI visibility tracking, talk to our SEO team.

FAQs

How many prompts should I be tracking for AI search visibility?

A minimum of 50 per site, covering ChatGPT, Google AI Overviews and Google AI Mode as a baseline. Track across more assistants if budget allows.

What’s the difference between mention rate and competitive win rate?

Mention rate is how often your brand appears across tracked prompts. Competitive win rate is how often you’re mentioned without a competitor appearing in the same answer. The second is a better indicator of genuine preference.

Can I use Google Search Console to find out what people are asking AI?

Not directly, but GSC’s question-style queries (filtered using a regex like what|where|how|why|when|should) are a useful proxy for the kinds of questions people are asking generally, including in AI tools.

Why don’t AI-driven enquiries show up properly in analytics?

Someone might see your brand mentioned in an AI answer, then search your brand name directly or call you without ever clicking a tracked link. The enquiry looks like direct or branded traffic, not an AI referral, even though AI is what put you on their radar.

How often should AI search visibility be reviewed?

Treat it like traditional rank tracking: ongoing, not one-off. AI referral patterns can shift sharply after a single model update, so monthly review is a sensible minimum.