How to Measure How Often AI Recommends Your Brand

How do I track whether AI recommends my brand?

You track it by measuring your named-in-answer rate. Build the set of questions your buyers ask AI when they shortlist suppliers, put each one to ChatGPT, Gemini and Perplexity, and record whether the answer names you or hands the buyer to someone else. Re-run the set monthly and watch the score move.

That number is now a sales metric. It maps to revenue you can win or lose before a buyer ever reaches your site. A buyer in your category opens a chat window and asks which provider is best for their situation. The answer it gives back becomes their starting shortlist, and that shortlist forms before anyone visits a website. If you are not in the answer, you are not in the conversation, and your analytics will never show the deal you lost.

Most teams measure rankings, traffic and impressions. None of those tell you whether AI is recommending you to a buyer who will never click. The named-in-answer rate does, and it is the one figure that maps directly to whether you make the shortlist.

The metric that now decides deals

In G2's buyer study surveyed in March 2026, 51% of B2B buyers now start their research with AI chatbots more often than with Google, and 69% chose a different vendor on the strength of AI guidance. Those exact figures are still being confirmed against G2's published release, but the direction is not in dispute. The shortlist is forming in the chat, and the question of whether you are named in it has a measurable answer.

Ranking and being recommended are different jobs with different metrics. You can hold position one in Google and still lose the buyer, because the AI summary sits above your listing and answers the question before the results load. Seer Interactive found in 2025 that organic click-through to the top result falls by around 60% once an AI answer sits above it. Pew Research Center found in 2025 that people click through to a website just 8% of the time when an AI summary is shown. A high ranking earns you far fewer visits once an answer sits above it.

So "our content already ranks fine" can be accurate and still leave you absent from the answer. A ranking tells you where you sit in a list of blue links. It says nothing about the AI summary now sitting above that list, or whether the summary says your name. The named-in-answer rate is the figure that does. The two move independently, so you can rank well and score zero, and plenty of strong-SEO companies do.

What named-in-answer rate actually counts

The named-in-answer rate is the percentage of your buyer-question set where an AI engine names your company in its answer. If you have 20 priority questions and three engines, that is 60 checks. Score each check as named or not named, and flag the ones where a competitor appears alongside you. The rate is the share that names you.

It is a share-of-voice figure for the answer layer. Track it as a single headline percentage, then break it down by engine, by question, and by which competitors surface next to you. The breakdown tells you where to act. The headline number tells you whether the programme is working.

How to build the question set that tracks your brand mentions in ChatGPT

The score is only as good as the questions. Build the set from how buyers actually phrase their problem, not from your product names or feature list. A buyer rarely asks for a brand. They ask AI to solve a situation, and the answer names whoever the model trusts on that situation.

Start with the questions that precede a purchase. "Best [category] for [buyer type]", "how do I solve [the pain your product removes]", "[your category] for [their regulated sector]", and the comparison and alternatives questions buyers ask once they have a shortlist. Twenty to forty questions covers most B2B categories at the start. Keep the wording verbatim, because phrasing changes the answer.

Group the set into three bands so the score means something:

Question band	What it captures	Why it matters
Branded	Questions naming you directly	Confirms the model knows who you are
Category	Unbranded "best provider for X" questions	Where the shortlist is actually decided
Problem	"How do I fix [pain]" questions	Earliest point a name enters the buyer's head

Category and problem questions are where deals are won or lost, because that is where a buyer with no shortlist gets handed one. A platform can be named in every branded question and zero category questions, which means the model knows it exists but never recommends it. That gap is the work.

Doing it manually versus running it as a programme

A one-off manual check gives you a baseline. The signal you actually want is how that figure moves across months, because a single reading tells you where you sit, not whether you are gaining or losing ground. AI answers shift when models update and when a rival starts getting cited in an answer you used to own.

You can run the first score yourself. Take your question set, put each question to each engine, log the result in a spreadsheet, and calculate the rate. It is honest, repeatable manual work, and it gives you a real baseline in an afternoon. The limits show up at the second score. Answers vary run to run, and engines update without notice. Holding a consistent method across months while doing your day job is where most internal attempts quietly stop.

Run as a programme, the score becomes a managed asset. Same question set, same method, scored every month, with the movement attributed question by question. You see the morning a rival surfaces in an answer your name used to own alone, while there is still time to respond. The monthly re-score is what turns a vanity check into a metric a sales leader can plan against.

What good looks like over time

A healthcare compliance platform we work with grew its AI-driven audience by 36% month on month while we ran this scorecard, and its largest contract that period came through ChatGPT. We anonymise the name by agreement, but the figures are confirmed and the mechanism is the point. The score moved because the work targeted the exact questions where the platform was absent, and the pipeline followed the score.

Set the bar against your own baseline, not an abstract target. If you start named in 4 of 12 priority category questions, the goal is 9 of 12, then hold it. The number that matters is the gap between where you are named and where your buyers are asking, and how fast you close it.

The scorecard as your first measurement

This scorecard is what we run for clients, and it is the deliverable, not a teaser for one. We are Forge Together, a B2B consultancy that moves companies from absent to recommended in the AI answers their buyers read. We call the discipline generative engine optimisation, run as a programme rather than a one-off project, and you can read what generative engine optimisation is in our guide to GEO. You can read the full method in the B2B GEO playbook, and how the citation side feeds the score in how to get cited in Google AI Overviews and AI answers.

The score is the start of a loop, not the whole of it. We audit where you stand, build the content the models draw on, earn the citations and structured signals that teach them to trust your name, then re-score every month to track the movement question by question.

The audit is your first scorecard, and it is free. We build your question set, run your category live across the engines, and show you your named-in-answer rate today, broken down by question and by competitor. You leave the call knowing exactly where AI names you today and which questions hand the buyer a rival instead.

Frequently asked questions

How do I track whether AI recommends my brand?

List the supplier-selection questions your buyers put to AI, run each one through ChatGPT, Gemini and Perplexity, and log whether the reply names you. The percentage that does is your named-in-answer rate. Re-run it monthly to track movement.

What is named-in-answer rate?

Named-in-answer rate is the percentage of your buyer-question set where an AI engine names your company in its answer. With 20 questions across three engines you have 60 checks. The share of those that name you is the rate. It is a share-of-voice metric for the AI answer layer.

Can I track brand mentions in ChatGPT myself?

You can run the first score manually by putting your question set to ChatGPT and logging each result in a spreadsheet. It gives you an honest baseline. The difficulty is consistency over time, because answers vary between runs and engines update without notice, which is why monthly scoring usually runs better as a managed programme.

Why isn't ranking on Google enough to measure this?

Ranking measures your position in a list of links. The named-in-answer rate measures whether the AI summary above those links says your name. Seer Interactive found in 2025 that click-through to the top result falls around 60% once an AI answer sits above it, so a high ranking no one scrolls to no longer guarantees the buyer reaches you.

How often should I measure AI brand recommendations?

Measure monthly. AI answers shift whenever a model updates or a competitor starts getting cited in territory you held. A single snapshot tells you where you stand; a monthly re-score tells you whether you are moving, which is the only evidence the work is paying back.

How many questions should my question set include?

Twenty to forty questions covers most B2B categories at the start, split across branded, category and problem bands. Category and problem questions matter most, because that is where a buyer with no shortlist gets handed one. Keep the wording verbatim, because how a question is phrased changes the answer the engine returns.

Find out where AI names you today

Book a discovery call and we will run your category live and show you exactly what AI says about you today. We build your question set, score your named-in-answer rate across the engines, and walk you through where you are named today and which questions still hand the buyer to someone else.

30 minutes. No pitch deck. Start with the free AI-visibility audit, and you leave with your first scorecard in hand.

Reviewed by the Forge Together GEO team.