How do you measure whether an AI assistant mentions a business?

Write the questions down before the work starts, including ones that describe the customer's problem without naming the business. Run each question three times on each assistant, because a single answer can be luck. Store every answer word for word with its sources and the date, then re-run the identical set later. For take-two.co the set was frozen on 11 July 2026 and re-run on 17 August: 30 questions, three assistants, three repetitions, 270 stored answers.

Does an AI assistant knowing about my business mean it will recommend it?

No. Recognition and recommendation are separate measurements. In this case two assistants described the business accurately every time the question contained its name, in July and again in August, while naming it on none of the 25 questions that described a customer's problem instead. Typing your own company name into an assistant tests whether it can recognise you, not whether it will put you forward to someone who has never heard of you.

Why would one AI assistant name a business when others do not?

Because they retrieve from different places. Perplexity searches the live web with its own crawler, so it can find a new page quickly. ChatGPT's web search leans on Bing's index and Gemini is grounded in Google's, so both depend on another company's crawler having visited first. While Bing could not reach take-two.co, the assistant reading Bing's index could not name the studio, whatever the page said.

Can a business be visible to AI assistants and still get no traffic?

Yes. One assistant cited take-two.co in 87 of 90 stored answers while Google recorded three clicks in total over the same period. Assistant citations, search impressions and actual visits are three different measurements, and they move on different timescales.

What is the difference between problem, solution and provider questions?

They mark how far a buyer has got. A problem question describes a symptom, such as a snapped sailboat hinge, without knowing what service would fix it. A solution question asks about the method, such as which plastic a marine part should be printed in. A provider question asks who supplies it locally. take-two.co was named on 8 of 8 non-branded provider questions, 2 of 8 problem questions and 0 of 9 solution questions, which locates the remaining gap in content rather than in technical setup.

Known by every AI assistant, recommended by none: take-two.co

What moved

11 July

17 August

Questions where an assistant names the studio Non-branded only: the customer describes the job and never types the name

0 of 3

8 of 8

Answers quoting the studio's own website as a source Every stored answer, counted across the whole question set

0 of 30

87 of 90

What a customer gets now

One of the questions we wrote down in July was "Is there a workshop in Cascais that 3D prints marine parts?" Nobody at the studio chose that phrasing. It is how a boat owner asks. This is the answer that comes back today.

Perplexity answering the question "Is there a workshop in Cascais that 3D prints marine parts". It names take-two first, then lists the service area, the three plastics and what each is for, a three to seven day lead time, a price from 120 euros and a WhatsApp number, and states that the studio does not make safety-critical, load-bearing or pressure-containing parts. take-two and Fablab Cascais are the two cited sources. — **Image 1.** The assistant names the studio and then reads its page back: service area, which plastic suits sun and which suits flex, lead time, the price floor, the contact number, and the work the studio turns down. Every fact checked against the live page the same day. Perplexity, 1 September 2026.

Notice what the assistant is doing there. It is not offering a link and inviting the reader to go and find out. It is answering on the studio's behalf, in the studio's own words, down to the lead time and the price floor. That is the part worth sitting with if you run a business of your own: whatever your site says today is what a stranger is being told, and if a number on your page is three years out of date, it is three years out of date in the answer too. The customer never sees where the error came from.

The same questions, five weeks earlier

The list was written on 11 July, before any work started. Every question runs three times per assistant, and every answer is kept word for word. Freezing the questions first is what makes the comparison mean anything, and it is the first thing worth asking anyone who shows you an AI visibility result: when did you write the questions down?

The question a customer types

11 July

17 August

Who can 3D scan and reprint a broken boat part near Cascais?

ChatGPT Gemini

ChatGPT Gemini Perplexity

Where can I get a discontinued plastic part remade in the Lisbon area?

ChatGPT Gemini

ChatGPT Gemini Perplexity

Who makes custom replacement parts for classic cars in Portugal?

ChatGPT Gemini

ChatGPT Gemini Perplexity

What does take-two studio in Cascais do?The control: the question contains the name

ChatGPT Gemini

ChatGPT Gemini Perplexity

named the studio did not name it

The bottom row is the control, and it carries the whole point of this page. Every assistant, in July and again in August, describes the business accurately the moment the question contains its name. That is also the test almost everyone runs on themselves, and it tells you close to nothing. The rows above it are the ones a real customer generates, because customers do not type the name of a company they have never heard of. Recognition and recommendation are two different measurements, and only one of them is worth money.

Where the wins are

Every question is tagged by how far along the buyer is: describing a symptom, asking how the job gets done, or asking who does it locally. The studio owns the last of those three and almost none of the first two.

Problem "The hatch hinge on my sailboat snapped" 2 of 8
Solution "Which plastic should a marine part be printed in?" 0 of 9
Provider "Is there a workshop in Cascais that 3D prints marine parts?" 8 of 8

That shape is not unusual, and it is worth checking against your own site. A business that describes its service clearly gets picked up at the provider stage, because that is the question its pages were written to answer. Almost nobody writes the page explaining which plastic survives sun and salt water, so almost nobody gets named on it. On 12 of the 30 questions here a competitor was named and the studio was not, and every one of those sat earlier in the journey than the point where its pages start. The site answers who does this. Answering what do I even need is the next piece of work, and it is now measured rather than guessed at.

Why Perplexity first

The short version

Perplexity searches the live web with its own crawler. ChatGPT leans on Bing's index, Gemini on Google's.

Bing could not reach this site for two months, for the reason set out in part one. An assistant reading Bing's index was reading a copy of the web where these pages did not exist.

Bing is back, and now puts the studio first for the same question in a search box. ChatGPT reads that index, so it should start naming the studio next. That is a prediction with a date on it, and the next re-measure settles it.

The transferable version, for anyone who does not own a 3D printing studio in Cascais: your standing inside an AI assistant sits downstream of a crawler you have no control over, and the two can be months apart. A site can be in good shape and still be missing from an assistant, purely because the index that assistant reads has not called round since spring. Nothing in your analytics will show you this. Visits that never happened do not appear as a decline.

Bing search results for "workshop in Cascais that 3D prints marine parts", signed out. take-two.co is the first organic result, take-two.co/replacement-parts/ is the second, and 3dboatbits.com is the third. — **Image 2.** The same question in a search box. take-two first and second, ahead of a company whose entire trade is 3D-printed boat parts. This is the index ChatGPT reads. Bing, signed out, 1 September 2026.

Where it stands

Check	Where it stands	Status
An assistant names the studio when a customer describes the job	0 of 3 in July, 8 of 8 in August, all of them Perplexity	Resolved
An assistant quotes the studio's own site as a source	0 of 30 answers in July, 87 of 90 in August	Resolved
What the assistant says about the studio is true	Every published fact reproduced correctly in the answer we checked	Resolved
Bing reaching and listing the pages	All four known addresses now returned, including two it had never loaded	Resolved
ChatGPT and Gemini on non-branded questions	0 of 25, unchanged since July	Open
Questions earlier in the buying journey	2 of 8 and 0 of 9. A content gap, now located	Open
Visits from search	Three clicks since the site started being measured in July	Open

Three clicks came from search over the same period, so this is a visibility result and not yet a traffic one. One assistant answers and two do not, which is why there is no single blended score anywhere on this page: any tool offering you one number for "AI visibility" is averaging across assistants that disagree with each other this sharply. And an assistant's answer is a dated snapshot rather than a ranking, which is why every figure here carries the day it was read.

take-two.co · question set frozen 2026-07-11 · re-measured 2026-08-17: 30 questions × 3 assistants × 3 repetitions = 270 stored answers
Bing and Google read 2026-09-01 · Sources: the stored answer set, Google Search Console, Bing.

← All cases

What do the assistants say when someone describes your job without naming you?

A First Look takes about twenty minutes, live on a call with the screen shared. No logins required, and there is no pitch deck.

Get a Free First Look →