How do we find out whether ChatGPT is recommending our firm?

Somewhere in your CRM there is probably a note that says "found us via ChatGPT". Maybe a new client mentioned it on the first call. Maybe your sales director heard it from a prospect who went elsewhere, because the assistant named three firms and you were not one of them. Either way, buyers are asking AI assistants who to hire, and you do not know how often you are in the answer.

They are asking in numbers. In G2's April 2026 survey of 1,076 software buyers, 51% said they now start their research in an AI chatbot, up from 29% a year earlier. That is a survey of software buyers, but the wider picture is similar: Forrester found in January 2026 that 94% of business buyers use AI somewhere in their buying process.

With Google you could see where you ranked. Nothing tells you how often ChatGPT, Gemini, Copilot or Perplexity recommended you last month, or who was named instead. So most founders ask ChatGPT once, then relax or panic, and a single answer justifies neither. Below is a free 15-minute check you can run today, then the repeatable method we use for tracking.

#Can we check what ChatGPT says about us in 15 minutes?

Yes. Open ChatGPT in a Temporary Chat set to unpersonalised, which does not use memory or custom instructions, and paste five questions a buyer might ask. Then run the same five in Google (looking for an AI Overview), Copilot and Gemini.

For example, a managed IT firm in the Midlands would try these, and you would swap in your own service, client type and region:

1. Shortlist: "Which managed IT providers would you recommend for a 100-person manufacturer in the West Midlands?"
2. Problem: "How do we switch IT support provider without downtime?"
3. Comparison: "Outsourced IT support or an in-house IT manager for a 100-person firm?"
4. Reputation: your firm's name followed by "any good for manufacturers?"
5. Head to head: your closest competitor's name, "or", your name, then "for a 100-person manufacturer?"

A good answer names you in the shortlist questions, describes what you do accurately and in current terms, gives a reason that matches the buyer's need, and cites a source you recognise, ideally independent coverage as well as your own site. Warning signs are hedges such as "according to its website", an old service or office, or a competitor named with a reason you could equally claim. Fifteen minutes will not give you a number you can trust, but it will tell you whether the full method is urgent.

#Why does a single check tell us so little?

Because AI answers change from one run to the next. Ask ChatGPT the same question three times and you may get three different shortlists: each answer is written fresh, with an element of chance in the wording and sometimes in the firms named.

Several other things move the answer:

• Who is asking. Assistants may use saved memory, past conversations and account settings, so your own account can lean towards your firm.
• Where they are. OpenAI says ChatGPT search "may use an approximate location based on your IP address to provide relevant local results".
• Which assistant. ChatGPT, Google's AI features, Copilot, Gemini, Perplexity and Claude draw on different sources and often name different firms.
• Whether it searched. For some questions the assistant searches the web live; for others it answers from what the model learned in training.
• Model updates and small changes in wording, either of which can change the list.

None of this makes tracking pointless. It means measuring the way a pollster does: the same questions, asked several times, in a controlled way. Patterns across many answers are reliable even when single answers are not.

#What should we measure?

Measure how often you are named, how you are described and why. That comes down to six things for each question and assistant:

• Named or not. Did the answer mention your firm at all?
• Prominence. Main recommendation, one of a shortlist, or a passing mention?
• Description. Accurate, current and specific, or vague and hedged?
• Competitors. Which other firms were named, and in what order?
• Sources. Which websites, articles and directories were cited or linked?
• Searched or not. Did the assistant search the web for this answer? That tells you whether new coverage can change it quickly.

The headline number is your share of answers: out of all the answers you collected, the proportion that named you. For example, ask 30 questions three times each in four assistants and you have 360 answers. If you are named in 36 of them, you appear in 10% of answers. Track that figure monthly, alongside the same figure for your main competitors.

#How do we set up tracking?

Setting up takes an afternoon; running it takes a few hours a month. ID-Pal is measured this way: "The same buyer questions run every month in ChatGPT, Gemini and Claude, in a fresh chat. A win only counts when it repeats, and losses are reported too."

#Step 1: Write the questions your buyers actually ask

Take them from sales calls and enquiry emails, in the buyer's words rather than your marketing language. Mix shortlist, problem, comparison and requirement questions ("which IT firms can help us get Cyber Essentials Plus?"), and add a few reputation questions that do include your name. Keep to a fixed list of 20 to 50: a short list that gets run beats a long one nobody runs.

#Step 2: Choose the assistants your buyers use

Start with ChatGPT and Google's AI Overviews and AI Mode, then add Microsoft Copilot and Gemini, then Perplexity and Claude. Do not skip Copilot if your buyers work in Microsoft 365: Forrester found business buyers four times as likely as consumers to use it. Our guide to which AI assistants buyers actually use explains how they differ.

#Step 3: Ask in a clean, consistent way

Set up a dedicated account for tracking, used for nothing else, with memory and custom instructions switched off, and run it from a UK connection. In ChatGPT, use a Temporary Chat set to unpersonalised. We do not rely on logged-out sessions: what a logged-out visitor sees may differ from what a signed-in buyer sees, and you cannot always tell which model answered. Paste each question exactly as written, and note the date, the assistant and the model name shown in the app. For Google, search in a private window and note whether an AI Overview appeared at all.

#Step 4: Record every answer in one sheet

Use one row per answer. These columns are enough, and you can copy them straight into a spreadsheet:

Save the full answer text alongside it. Your share of answers is then =COUNTIF(E:E,"Yes")/(COUNTA(E:E)-1), where column E is "Named" and the minus one skips the header row. Count a question as won only when you are named in at least two of its three runs.

#Step 5: Run a baseline

Ask every question in every assistant, three times each, before you change anything. If you appear once in three runs, you are on the edge of the shortlist. Without a baseline you cannot show progress later.

#Step 6: Repeat on a fixed rhythm

Run the list several times a month, for example once a week, and report monthly. When a major model update is announced, note it in the sheet so you can explain sudden shifts.

#Which tools track AI recommendations, and what do they cost?

Several tools now ask your questions automatically, repeat them and record sources. They save hours once the manual method becomes a chore, but choose one that lets you enter your own buyer questions, shows the actual answers rather than just a score, and exports the data.

Prices change often, so check each vendor's current page. Most price by how many questions and assistants you track: Otterly's plans, for instance, differ mainly by the number of prompts included. Compare the cost per tracked question, and check a sample of the tool's answers by hand. Our guide to building a simple AI visibility dashboard shows how to report what the tool collects.

#What do GA4 and Search Console show now?

Part of the picture, for free. Since 3 June 2026 Search Console has generative AI performance reports showing impressions for AI Overviews and AI Mode, rolled out to all sites by 31 August 2026. Since 13 May 2026 GA4 has had a default AI Assistant channel for visits from ChatGPT, Gemini, DeepSeek, Copilot and Grok; Claude and Perplexity are not on its list, and clicks from AI Overviews and AI Mode still count as Organic Search.

Neither report tells you how often you were named in an answer that did not link to you, or who was named instead. That is why the question list still matters.

#How do we read the results?

Read for patterns, not single answers. Each pattern points to a different fix.

You are known by name but not recommended. Assistants describe you correctly when asked directly, but leave you out of shortlists. ID-Pal was in exactly this position: "AI could describe ID-Pal. It never recommended it." Its case study adds: "Across 36 recorded AI answers to the questions ID-Pal's buyers actually ask, ID-Pal was named three times, and only for its brand or its Salesforce integration." After pages were rebuilt around buyer questions: "UK buyer questions where AI names ID-Pal: none in May, two in August, four in September 2026." The fix is usually pages that describe you in the buyer's terms, plus independent mentions that confirm them.

You are named only in a narrow niche. You appear for "IT support for dental practices in Coventry" but not "IT support in Coventry". That is fine if the niche is your market; if not, you need content and coverage in the wider category.

You are named but described wrongly. Out-of-date services, an old office, the wrong headcount. Our guide to fixing wrong AI descriptions of your firm explains how.

Competitors are cited from sources you are missing from. If the same trade articles, directories or comparison pages keep appearing, those are the places you need to be.

Answers hedge about you. "According to the company" suggests the assistant found your claims only on your own website, which points to a need for independent coverage.

The same sheet gives you a competitor view, because you have already recorded every firm named. Our guide to measuring AI share of voice against competitors turns it into a benchmark.

#How do we connect tracking to enquiries?

Tracking tells you whether you are named, not whether it brings in work. For that, ask the people who get in touch.

• Add "an AI assistant such as ChatGPT" to the "how did you hear about us?" field on your enquiry form, with a free-text box.
• Ask the same question on every first call and record the answer in your CRM.
• Watch the GA4 AI Assistant channel, and add Claude and Perplexity referrals yourself.

Expect this to understate the effect: many buyers read an AI answer, then search your name or type your address, so the visit looks like ordinary search or direct traffic. When the channel works, it shows in the numbers: at Helm Club, "1 in 5 new members this year came through an AI recommendation." Our guide to tracking leads in the AI era covers this in more depth.

#What to do this month

1. Run the 15-minute check above in ChatGPT, Google, Copilot and Gemini.
2. Ask your sales team for the questions prospects most often ask before they buy, and turn them into a fixed list of 20 to 50.
3. Set up a dedicated UK tracking account with memory off, and copy the sheet headers above.
4. Run your baseline, three times per question per assistant, recording the model each time.
5. Add an AI assistant option to your enquiry form and first-call notes, and check the GA4 AI Assistant channel.
6. Book the weekly runs and a monthly review, and share the baseline with your fellow directors.

#Frequently asked questions

#Is there a free tool to check if ChatGPT mentions my brand?

ChatGPT itself, used in a clean session with a spreadsheet, is the free tool, and it is how we suggest every firm starts. Google Search Console and GA4 are free and show part of the picture for your own site. Some paid trackers offer trials: Otterly advertises a seven-day free trial with no card needed.

#Does ChatGPT give different recommendations in the UK and the US?

It can. OpenAI says ChatGPT search may use an approximate location based on your IP address to give relevant local results, so the same question asked from London and New York can produce different firms. Track from a UK connection, and include the country or region in questions where your buyers would.

#How often do ChatGPT's recommendations change?

Every answer can vary, which is why you ask several times. Assistants that search the web live can reflect a new page or article within weeks; answers drawn from a model's memory change only when the model is retrained, which takes months. ID-Pal's record shows the pace: "Claude names it for UK fund administrators, on a page rebuilt in August, and links NorthRow to ID-Pal in the same answer."

#What if we are never named at all?

That is common, and it is a starting point rather than a verdict. Look at who is named and which sources are cited for them, check that your website says plainly what you do, for whom and where, and that AI crawlers can read it. Then earn independent mentions in the places the assistants already cite.

#Do we need to track Microsoft Copilot as well?

Yes, if your buyers work in organisations that use Microsoft 365, because Copilot sits inside the tools they use every day. Start with ChatGPT and Google, add Copilot and Gemini, then Perplexity and Claude.

If you want a first look before you build the sheet, book a free Growth Review: 30 minutes with David Adams, with your site, two competitors and what Google and ChatGPT say about all three checked beforehand, and three fixes in writing afterwards.

#Sources