Why measure at all
When a customer asks an AI assistant who to call, the answer names a few businesses, and that shortlist is often where the customer starts. If you’re not on it, there’s no lost click in your analytics and no ranking to check. Without measuring, you can’t tell whether you have a gap, how big it is, or whether anything you change makes a difference.
A careful measurement gives you three things:
- Where you stand today: how often you’re named, and who is named instead.
- Priority gaps: which public information behind the answers looks missing, wrong or inconsistent — as findings, not a DIY brief.
- A fair baseline: a like-for-like starting point to re-check against after fixes.
The rest of this page explains how we keep that measurement honest, so the numbers mean something when you act on them.
What we measure
One thing, done carefully: when someone asks an AI assistant the kind of question your customers ask, does the answer name your business?
For every answer we record one of four outcomes:
- Named your business appears in the answer.
- Cited your website or listing is used as a source, but your business isn’t named in the answer.
- Not named the assistant answered, and your business wasn’t in it.
- No answer the assistant refused, failed, or gave no usable answer.
We also note which other businesses were named, the order they appeared in, the sources the answer cited, and anything the assistant said about you that is out of date or wrong.
The assistants we test
We test the three assistants UK customers are most likely to use for this kind of question:
- ChatGPT (OpenAI), with web search on
- Gemini (Google)
- Microsoft Copilot
Every result records the engine, the mode used and the date. We run questions in a fresh or temporary chat with no memory or history, and state the town or area in the question, so earlier conversations don’t colour the answer. Glowing Terms is independent and has no affiliation or partnership with OpenAI, Google or Microsoft.
The question set
We agree a short set of realistic customer questions with you before we start, written the way people actually ask, for example:
- a general “who’s best” question: “Who are the best loft conversion companies in Leeds?”
- a local-area question: “Loft conversion near Headingley with good reviews?”
- a price-led question: “How much does a dormer loft conversion cost in Leeds, and which companies would you recommend?”
- a trust-led question: “TrustMark-registered loft builders in Leeds?”
The exact wording of every question goes in your report. When we re-measure, we use the same questions, word for word, on the same engines, so a change in the count reflects a change in the answers rather than a change in what we asked.
Repeated runs
AI assistants don’t give the same answer every time. A single screenshot can make a business look invisible or dominant by chance. So we put each question to each engine more than once, and record every run separately. The number of runs per question is stated in the report.
A typical snapshot might be 5 questions × 3 engines × 2 runs = 30 runs. Larger audits use more runs.
Denominators, always
Every count comes with what it is out of. Not “you’re named 4 times”, but “named in 4 of 28 answered runs”. You’ll see the breakdown per engine and per question, so you can check the totals yourself.
| Illustrative snapshot (fictional) | Runs |
|---|---|
| Total runs (5 questions × 3 engines × 2 runs) | 30 |
| Runs that returned no answer, listed separately | 2 |
| Answered runs (the denominator) | 28 |
| Answered runs that named the business | 4 of 28 |
Illustrative example with a fictional business, matching the sample report on our home page. Not real results.
No-answer runs are listed separately
Sometimes an assistant declines to recommend anyone, errors, or gives an answer with no businesses in it. Those runs aren’t “not named”, and they aren’t quietly dropped either. We list them on their own line, with the engine and question, and leave them out of the denominator. That way the headline count is never flattered or dragged down by runs that didn’t really answer.
Public information only
Our audits use only public information about your business (your website, Google Business Profile, directory and review listings, and similar sources) and the answers the assistants give. We never ask for or use your customers’ or patients’ personal data, and we don’t put named individuals’ personal data into AI tools. See our privacy notice.
Fix, then re-measure the same way
The snapshot flags priority gaps as findings (for example thin or inconsistent public profiles, missing service or area coverage, or crawler-access issues) — not a step-by-step DIY brief. The Fix pack is where we correct the public information or hand you ready-to-paste changes, then re-run the same question set on the same engines and give you a plain before/after sheet with dates, run counts and denominators. See how it works and pricing.
We don’t credit our work for changes in your enquiries or revenue. The before/after sheet reports what the assistants said, nothing more.
What we don’t promise
AI answers vary. Assistants don’t give the same answer every time, and results can change between runs, days and users. Every result we report applies only to the specific questions, engines and dates tested. Nobody can guarantee that ChatGPT, Gemini, Copilot or any other AI assistant will name, rank or recommend a business, and we don’t.
For clinics, our work covers online visibility and factual listing data only. We make no assessment of clinical quality.
Questions about the method? Read the FAQ or email hello@glowingterms.co.uk.