On 19 October 2026 Manual Focus launched Manual Focus Light Meter, the instrument behind its AI Shortlist research. It asks six AI engines the questions UK buyers ask and measures which brands they name, with a 95% range, and which sites they cite. Asked one question three times, an engine named the same brands only 20% of the time.
See what Light Meter measures, open the indoor cycling report, or, if your brand is in that study, request your free AI Shortlist card.
Key takeaways
- Manual Focus Light Meter measures six engines separately, ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode, from a UK location.
- Every question runs three times per engine, because in the indoor cycling study an engine gave exactly the same brands across all three runs only 20% of the time.
- Every rate carries a 95% range, so Wahoo’s 83% share of UK smart trainer answers reads as 79% to 87%.
- The engines split on challengers, with Van Rysel named in 43% of Claude’s trainer answers and 5% of Google AI Mode’s.
- Editorial sites supplied 51% to 56% of citations in indoor cycling, against 3% to 5% for Reddit, YouTube and forums.
What is Manual Focus Light Meter?
We built Light Meter to answer one question for a brand team, whether your brand makes the shortlist when a UK buyer asks AI what to buy. We write 100 to 300 questions the way buyers ask them, tag each with the buying moment behind it, from value and beginner to performance and problem questions, then put every question to six engines from a UK location, three times each. The answers are collected through DataForSEO.
Light Meter records every brand named and every site cited, then works out each brand’s shortlist rate, the share of answers that name it, with a 95% range. The first study, which indoor cycling brands AI recommends in the UK, used 309 questions and produced 5,562 answers. The full interactive report is open to everyone.
How is it different from rank tracking?
A rank tracker tells you where a page sits in a list of results for a keyword. In an AI answer there’s no list to sit in, only a short set of brands the engine chose to name and the sources it leaned on. We argued in June that the citation moved off the page, and Light Meter is our attempt to measure where it went.
| Question | Rank tracker | Manual Focus Light Meter |
|---|---|---|
| What it counts | Your page’s position for a keyword | Whether an AI answer names your brand |
| Unit of measure | One position per keyword | The share of answers that name you |
| Repeat runs | One reading per keyword per check | Three runs of every question per engine |
| Engines | One search engine at a time | Six AI engines, reported separately |
| Uncertainty | None shown | A 95% range on every rate |
| Questions | Keywords you choose | Unbranded buyer questions, tagged by buying moment |
| Sources | The ranking URL | Every site the engine cites, by type |
Why ask the same question three times?
Because AI engines don’t give the same answer twice. In the indoor cycling study, an engine named exactly the same set of brands across all three runs of a question only 20% of the time.
| Engine | Same brands in all three runs |
|---|---|
| Perplexity | 48% |
| Claude | 20% |
| ChatGPT | 13% |
| Google AI Overviews | 12% |
| Gemini | 11% |
| Google AI Mode | 11% |
So one check of what ChatGPT says about your category is one draw from a moving set. A SparkToro study reported by Search Engine Journal in January 2026 gave brands the same advice, to track how often they appear, not where.
Why measure each engine separately?
At the top of a category the engines agree. All six put Wahoo first for smart trainers and Zwift first for training apps. Below the leaders they behave like different shops with different buyers. Van Rysel, Decathlon’s own brand, appeared in 43% of Claude’s trainer answers and 5% of Google AI Mode’s. JetBlack ranged from 47% on Claude to 11% on AI Mode, and Saris showed up in 40% of Claude’s trainer answers and under 10% of every other engine’s.
Part of that gap is how each engine answers. Claude named 4.2 brands per answer on average in the study, against 2.9 to 3.3 for the other engines. Google AI Overviews appeared on only 64% of the searches, which pulled down its rate for most leading brands.
Our view at Manual Focus is that one blended “AI visibility” score hides the most useful finding for a challenger brand, which is the engine where it’s already close and the one where it’s missing.
Why does every number carry a range?
A shortlist rate is an estimate from a sample of answers, and Light Meter shows its margin as a 95% range, which tells you whether a gap between two brands is real or could be noise from the sample.
In indoor cycling, Wahoo’s smart trainer rate of 83% has a range of 79% to 87%, well clear of Tacx at 65%. Rouvy at 71% and MyWhoosh at 69% are a different story. Their ranges overlap, and the report gives both a rank range of second to third, so calling either one second would claim more than the data shows. Wahoo’s 64% against Wattbike’s 58% for smart bikes is closer than it looks, because those two ranges overlap as well. A brand team that knows this saves itself from spending a quarter’s budget closing a gap that may not exist.
Why track the sites AI cites?
The sources tell you where to do the work. In indoor cycling, editorial sites supplied 56% of citations for trainers, 53% for smart bikes and 51% for apps, led by BikeRadar, Cyclingnews and Cycling Weekly. Reddit, YouTube and forums supplied 3% to 5%. Brand websites mattered most for apps, at 30% of citations.
For most brands in that category the job is earned media, getting into the buyer’s guides the engines quote, more than rewriting their own pages. The earned media pitch generator and our piece on earned channels when journalists also use AI cover how to go after those placements.
What can you order?
| Offer | What you get | Price |
|---|---|---|
| AI Shortlist card | A one-page card for a brand already in a study, with its rate on each engine, the buying moments it wins and loses, and the sites that cite its rivals but not it | Free |
| Category audit | A full AI Shortlist study of your category, built to order in five working days | On application |
| Brand Snapshot | Your brand against up to five competitors on the questions that matter to you, with source gaps and a short list of actions | On application |
| Fix programme | We measure, work on the pages and publications that decide your category, then measure again to show what moved | On application |
Brands in the indoor cycling study can request their free card now. The next published edition covers trail running kit, shoes, waterproof jackets, hydration packs and races, in January 2027. The Light Meter page lists the other categories we can build to order.
Where should a brand team start?
- Write down the ten unbranded questions your buyers ask most, and tag each with its buying moment. The category entry points playbook is the method.
- Put those questions to at least three engines, three times each, and count how often you’re named on each one before you draw any conclusion.
- List the sites the engines cite for your category, then check whether you’re in them and whether the details are current.
- Set a date three months out to run the same questions again, so you can tell whether anything you changed moved the rate.
What Light Meter can’t tell you
Light Meter counts whether an engine names a brand, not whether it praises it, and not whether anyone buys. Logged-in users with chat history may see different answers. Perplexity and Claude are measured through their APIs, which can differ from their apps. Each study is a snapshot of the month it ran, so a category needs repeat editions before it shows a trend.
Disclosure. James Vickers has worked with Wahoo and Wattbike, was on the founding team at RGT Cycling, and currently works with Ribble Cycles. Category audits run on the same method whoever orders them, and we don’t change results on request.
Further reading
- Which indoor cycling brands does AI recommend in the UK?, the first study in full.
- The citation moved off the page, on why rankings stopped predicting who AI quotes.
- AI Overviews hit 82% of tracked brand searches, on how often AI now answers a search for your own name.
- SEO playbook for premium endurance brands, the search groundwork behind AI visibility.
- Category entry points, for mapping the buying moments in your own market.
If you’d like Light Meter run on your category, tell us what you sell.
Frequently asked questions
What is Manual Focus Light Meter?
Manual Focus Light Meter is the measurement tool behind the Manual Focus AI Shortlist research. It asks six AI engines, ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode, the questions UK buyers ask, three times each from a UK location, and records which brands they name, with a 95% range on each rate, and which websites they cite.
How do you measure a brand's visibility in AI answers?
Ask the unbranded questions your buyers ask, put each one to every AI engine you care about several times, and count the share of answers that name your brand. Manual Focus Light Meter runs each question three times on six engines and gives every rate a 95% range, because one answer from one engine is too unstable to plan around.
Why do AI answers change when you ask the same question twice?
AI engines generate a fresh answer each time, so the brands they name shift between runs. In the Manual Focus indoor cycling study, an engine named exactly the same brands in all three runs of a question only 20% of the time, from 11% for Gemini and Google AI Mode to 48% for Perplexity.
Is an AI visibility tool the same as a rank tracker?
No. A rank tracker records where your page sits in a list of search results for a keyword. An AI visibility measure such as Manual Focus Light Meter records whether an AI engine names your brand in its answer, how often across repeated runs, on which engine, and which websites the engine cites as its sources.
How much does a Manual Focus Light Meter audit cost?
Manual Focus prices its Light Meter category audits, Brand Snapshots and Fix programme on application. Category audits are built to order in five working days. Brands already in a published study, such as UK indoor cycling, can request a free one-page AI Shortlist card through the Manual Focus enquiry form.
Sources
- Light Meter, AI visibility measurement for brands, Manual Focus
- AI Shortlist research, UK indoor cycling, full interactive report, Manual Focus
- Which indoor cycling brands does AI recommend in the UK?, Manual Focus
- AI recommendations change with nearly every query, SparkToro study, Search Engine Journal
- AI Optimization API documentation, DataForSEO
From the podcast
James and Hannah introduce Manual Focus Light Meter, the tool behind the Manual Focus AI Shortlist research, and explain why measuring AI visibility needs three things a rank tracker doesn’t do. Repeated runs, because in the UK indoor cycling study an engine named the same brands across three runs of a question only 20% of the time. Separate engines, because Van Rysel appeared in 43% of Claude’s trainer answers and 5% of Google AI Mode’s. And 95% ranges, because Rouvy and MyWhoosh look like second and third but overlap. They finish with three jobs to hand to AI, what Light Meter can’t tell you, and the next edition, trail running kit in January 2027.
Get AI on this
Paste these into whichever AI assistant you use: ChatGPT, Claude, Gemini or Copilot. Swap the bracketed parts for your own details. The first needs file uploads. The second needs an assistant that can run searches on other engines and a schedule, or a person to paste the answers in. The third needs web browsing.
1. Write your question set
I run marketing for [brand], which sells [category] in the UK.
I want the 10 unbranded questions buyers ask most before they choose a brand
in my category, so I can check which brands AI engines name in reply.
Attached are:
- a Search Console export of queries for the last 3 months
- our customer service tags or ticket subjects for the same period
Write the 10 questions the way a UK buyer would type them into an AI
assistant, with no brand names in them. Tag each with its buying moment:
value, beginner, performance or problem (something isn't working).
Give me a table with the columns: Question | Buying moment | Evidence (the
queries or tags it came from).
Don't invent search volumes or demand. If the files don't support a
question, leave it out and say why.2. Run the questions on a schedule
Set this up as a routine that runs once a quarter, on the same questions
each time.
Here are 10 unbranded buyer questions for [category] in the UK:
[paste questions]
Ask each question three times on each of these AI engines, in a fresh
conversation each time and set to a UK location where possible: [list at
least three engines you can reach]. If you can't reach an engine yourself,
say so and give me a simple form so a person can paste the answers in.
Record every brand each answer names. Then give me a table with the columns:
Engine | Question | Brands named in run 1 | Run 2 | Run 3 | Times [brand] was
named (0 to 3) | Times [competitor 1], [competitor 2] and [competitor 3] were
named.
Finish with our shortlist rate on each engine (answers naming us divided by
all answers), the change since the last run, and a note that rates from 30
answers are rough, so small gaps between brands may be noise. List any runs
you couldn't complete.3. Find the sites the engines cite
I run marketing for [brand] in [category], UK.
For these buyer questions: [paste 5 of your questions], find the websites AI
engines cite when they answer. Where you can see an engine's cited sources,
use those. Otherwise find the buyer's guides, reviews and round-ups that rank
for the questions.
Give me a table with the columns: Site | Type (editorial, retailer, brand,
forum, video) | Guide title | URL | Date last updated | Is [brand] included? |
Who is listed instead of us.
Then give me a hit list of the five guides where getting included or
corrected would matter most, with the ones that leave us out at the top. If
you couldn't open a page, mark it "not checked" rather than guessing what it
says.Read the episode transcript
James: If you've ever typed your category into ChatGPT, seen your brand in the answer and felt relieved, I'd ask you to type it again. Today we're talking about the tool we built to measure that properly, Manual Focus Light Meter. I'm James Vickers, this is Manual Focus, and Hannah's here to check I don't oversell it.
Hannah: Hi James. I'll start with what it does. Light Meter takes the questions UK buyers ask, usually somewhere between a hundred and three hundred of them, and puts each one to six AI engines. ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode, all from a UK location, three times each.
James: And it counts what?
Hannah: Two things. Which brands each answer names, and which websites each engine cites as its sources. From that it works out a brand's shortlist rate, the share of answers that name it, and every rate comes with a ninety-five per cent range.
James: Before we go on, the same disclosure as in our indoor cycling episode. I've worked with Wahoo and Wattbike, I was on the founding team at RGT Cycling, and I currently work with Ribble Cycles. The audits run on the same method whoever orders them, and we don't change results on request.
Hannah: And the evidence for all of this is our first study, on UK indoor cycling. Three hundred and nine questions, five thousand five hundred and sixty-two answers.
James: So start with the obvious objection. Most marketing teams already have a rank tracker. Why isn't that enough?
Hannah: Because a rank tracker answers a different question. It tells you where your page sits in a list of results for a keyword. In an AI answer there's no list to sit in. There's a short set of brands the engine chose to name, and the sites it leaned on to choose them.
James: Which is the argument we made back in June, when we said the citation had moved off the page. Ranking well and being quoted by AI had come apart. This is us trying to measure where it went.
Hannah: Right. And the first thing you hit is that the answer won't sit still.
James: Go on.
Hannah: We asked every question three times on every engine. An engine named exactly the same set of brands across all three runs only twenty per cent of the time. Perplexity was the steadiest at forty-eight per cent. Gemini and Google AI Mode were down at eleven.
James: So the screenshot of ChatGPT in last month's board deck.
Hannah: Is one draw from a moving set. SparkToro gave brands the same advice in January, track how often you appear, not where.
James: Then the six engines. Why not blend them into one number? That's what a board wants.
Hannah: Because they disagree, and the disagreement is where the opportunity is. At the top they agree. All six put Wahoo first for smart trainers and Zwift first for apps. Below that, Van Rysel, Decathlon's own brand, appeared in forty-three per cent of Claude's trainer answers and five per cent of Google AI Mode's.
James: Our view at Manual Focus is that one blended score hides the most useful thing a challenger brand can know, which engine it's already close on and which one it's missing from.
Hannah: And part of that is just how the engines answer. Claude named about four brands per answer, the others nearer three.
James: Now the ranges. I suspect that's the bit people skip.
Hannah: It's the bit that saves money. A shortlist rate is an estimate from a sample of answers, so it has a margin. Wahoo's eighty-three per cent for smart trainers runs from seventy-nine to eighty-seven, well clear of Tacx at sixty-five. But Rouvy at seventy-one and MyWhoosh at sixty-nine overlap. You can't honestly call either one second.
James: So if you're MyWhoosh's marketing lead, you don't spend a quarter's budget closing a two-point gap that might not exist.
Hannah: Exactly. And the last thing it measures is the sources. In indoor cycling, editorial sites supplied more than half the citations, led by BikeRadar, Cyclingnews and Cycling Weekly. Reddit, YouTube and forums, three to five per cent.
James: Which tells you the work is mostly earned media. Getting into the buyer's guides the engines quote, more than rewriting your own pages. So how does a team start on this without commissioning a study? Get AI on this.
Hannah: What's first?
James: First, get your AI to write your question set. The ten unbranded questions your buyers ask most, each tagged with its buying moment, value, beginner, performance, problem.
Hannah: Give it your search queries and your customer service tags, and it hands back the ten questions, phrased the way a buyer would ask them, with the evidence for each one.
James: Second, have it run those questions on a schedule. Three engines at least, three times each, counting how often you're named.
Hannah: Set it as a quarterly routine, so the same questions run again in three months and you can see whether anything you changed moved the rate. Where your AI can't reach an engine, it should say so, and a person pastes the answers in.
James: Third, the sources. Have your AI list the sites the engines cite for your category and check whether you're in them.
Hannah: You get a hit list of guides, with the ones that leave you out at the top. Getting into them is still a human job.
James: The prompts are in the show notes.
Hannah: And what Light Meter can't tell you. It counts whether an engine names a brand, not whether it praises it, and not whether anyone buys. Logged-in users with chat history may see different answers. Perplexity and Claude are measured through their APIs, which can differ from the apps. And each study is a snapshot of one month, so a category needs repeat editions before it shows a trend.
James: Which is why the next edition matters. Trail running kit, in January twenty twenty-seven.
Hannah: And if your brand is already in the indoor cycling study, there's a free one-page AI Shortlist card you can ask for.
James: So the takeaway. Measuring AI visibility means asking more than once, reading each engine on its own, and knowing which gaps are real before you act on them. The full article and the Light Meter page are on manual-focus dot co dot uk. If you want the next study when it lands, follow the show in Apple Podcasts, Spotify or wherever you listen. And if you'd like to work out what this means for your own brand, book a free working session with us on the site. Thanks, Hannah.
Hannah: Thanks, James. Until next time.