RecOpt

Find the question where AI assistants start recommending you

RecOpt asks ChatGPT, Gemini and Perplexity the way your buyers do, from broad questions to narrow ones. It shows the step where they start naming your brand, who they name before that, and what to change.

Measured on a live store: ChatGPT named it first, as the best match, only once buyers asked to try fonts on their own photo. One step earlier it named six other shops.

Where to buy decorative fonts

ChatGPT, 6 Nov 2026
  1. 01Where can I buy fonts?Named instead: MyFonts, Creative Market, FontspringNot named
  2. 02…decorative display fonts?Named instead: Creative Market, MyFonts, YouWorkForThemNot named
  3. 03…with a cheap one-time commercial license?Named instead: Taleggio Foundry, K-Type, Creative MarketNot named
  4. 04…and try them on my own photo first?Named first, marked “Best match”Named

Each step is a new chat that adds one buyer condition. The condition that first brings the brand in is the one to build content around.

How it works

A share of mentions tells you where you stand. A path tells you what to do.

Most tools count how often an assistant names you. That number does not say which sentence in a buyer’s question makes the difference. RecOpt finds it in three steps.

1

Ask like a buyer

One need, asked from broad to narrow. Each step adds a single condition, such as a budget, a licence type or a feature, and runs in a fresh chat so earlier answers cannot leak in.

2

See your path

You get the first step where an assistant names you, the condition that brought you in, and who and which websites the assistant used before that step.

3

Change one thing and re-measure

Log the change, ask the same questions again and compare with questions you left alone. If the difference is within noise, RecOpt says so.

Real example

Same store, three assistants, three different pictures

Letterlays sells decorative fonts and letter art worldwide. We measured it on 6 November 2026 with 17 buyer questions and four narrowing ladders.

3 of 3ChatGPT ladders named the store once the question asked to try a font on your own photo before buying. One step earlier: 0 of 3.
0 of 62Gemini answers named the store, although it describes the brand correctly when asked directly. A likely reason: the site is young and ranks weakly in Google, the search behind Gemini.
55 of 66Perplexity guest questions ended with a request to sign in, so that assistant is only partly measurable today.

What we did with it

9 Nov 2026
FindingAction
The page for “try on your own photo” was blocked from search.Made it an indexable page with the question as its title.
ChatGPT said the letter packs could not be bought. They could.Asked Google to recrawl the catalogue.
Roundups of font shops did not list the store.Drafted four honest pitches to their authors.

We re-measure on 3 December and report the change with intervals, including when it is zero.

What you get

Fewer numbers, more decisions

The step where you appear

For each need, the first buyer condition that makes an assistant name you, and the gain each condition adds over the step before.

Who and what stands in your place

The brands named instead of you and the websites the assistants cite, per assistant, since each one leans on different sources.

Honest statistics

Every rate comes with its interval and the number of answers behind it. Assistants are never averaged together.

A check on the search engine behind it

Gemini leans on Google, ChatGPT and Copilot on Bing. If an assistant knows your brand but never finds your site, the report says where to look first.

Covered today
Geminiguest, automatedPerplexityguest, partialChatGPTsigned out browser
Questions

What people ask before joining

What does RecOpt measure?

How often AI assistants name your brand when buyers ask about your category, from broad questions to narrow ones. For each question it records who is named instead of you and which websites the assistant cites, and it reports the step at which you start to appear.

How is this different from a visibility tracker?

A tracker shows a share of mentions. RecOpt asks the same need at several levels of detail, one added buyer condition per step, each in a new chat. That shows which condition makes an assistant name you, and so what to change on your site and elsewhere.

Which assistants do you cover?

Today: Gemini and Perplexity asked as a guest, and ChatGPT captured in a regular signed out browser session. Perplexity asks guests to sign in for part of the questions, so its coverage is partial. More assistants follow as their terms allow.

Is asking assistants through their websites allowed?

We ask the way a person does: one question at a time, about one every 20 seconds, never signing in on someone else’s behalf and never solving robot checks. Each assistant has its own terms. We review them per assistant before the paid launch and drop an assistant if its terms do not allow it.

How many answers do I need to trust a number?

One answer means very little. With about seven repeats per question the error is under 16 points. RecOpt always shows an interval and the number of answers next to a rate, and compares assistants separately instead of averaging them.

Can RecOpt prove that my change worked?

It records the change in a log, then re-measures the same questions and compares them with questions you left alone. When the intervals overlap it says there is no measurable change yet.

What data do you store?

On this site only the email address you give to join early access. There are no cookies and no third party trackers. You can remove your address on the privacy page.

When does it launch and what does it cost?

Early access comes first, with a free check of one brand. Prices are not set yet. Join the list and we email once when your spot opens.

Know which question brings you in

Join early access and get a free check of one brand when it opens.