Why does ChatGPT sound unsure about our firm, and how do we fix it?

Ask ChatGPT about your firm and read the tone of the answer. For some firms it is crisp, for example: "Example Partners is an employment law firm in Leeds that advises hospitality and retail groups." For others it hedges: "Example Ltd appears to offer consultancy services, according to its website." Sometimes it says it has little information, or names you last in a list with a description that could fit anyone.

That tone is not an accident. AI assistants such as ChatGPT, Gemini, Claude, Copilot and Perplexity state things plainly when several independent sources agree, and hedge when the main evidence is your own website. Independent evidence is most of what they cite: in Muck Rack's May 2026 analysis of more than 25 million links cited by ChatGPT, Claude and Gemini, earned media made up 84% and paid content 0.3%.

Buyers notice the hedge. In the TrustRadius 2026 B2B Buying Disconnect Report, 63% of technology buyers had used AI in a purchase, and 94% of those checked what it told them at least some of the time. A buyer who reads "appears to" goes looking for a firm the assistant is sure about. If the problem is facts that are plainly wrong rather than vague, start with how to fix wrong facts ChatGPT repeats about your firm.

#What does it mean when ChatGPT hedges about our firm?

It means the assistant has not found enough independent evidence to state your claims as fact. No assistant shows a confidence number, but the signs are consistent:

• phrases such as "appears to", "seems to", "according to its website", "describes itself as" or "claims to specialise in"
• a generic description that could fit any firm in your sector
• out-of-date details, such as an old office or a service you stopped offering
• your firm missing from shortlists where competitors appear, or named last with less detail
• an admission that it has limited information about you

Each tells you the same thing. Confident answers state facts; hedged answers attribute them. If a claim appears on your website and nowhere else, the assistant tends to credit it to you. If your site, a trade article and your professional body's register all say it, the assistant can say it plainly. For a buyer, the difference between "a specialist in X" and "appears to offer services" is the difference between a firm worth calling and one worth skipping.

#Why do assistants hedge about some firms and not others?

Because the evidence about some firms is thin, comes only from them, or disagrees. That is how a careful buyer behaves too.

Evidence reaches an assistant by two routes. The first is training: the model has read a large snapshot of the public web, and a firm described the same way in many places leaves a firmer impression than one that appears mainly on its own site. The second is live search. Gemini, Google's AI Overviews and AI Mode start from Google's index; Copilot starts from Bing's; and ChatGPT search has drawn on Bing, with 87% of its citations matching Bing's top results in a small Seer Interactive test in February 2025. So the index matters: a firm that is well described in Google but patchy in Bing can get a confident answer from Gemini and a vague one from Copilot.

On both routes, your website is necessary but not enough. It is where the assistant finds the facts; other sources are where it learns whether anyone else agrees. Independent sources carry weight with buyers as well: in G2's April 2026 survey of 1,076 software buyers, 45% said citations of review sites were the most confidence-inspiring signal in an AI answer. Why earned coverage counts for more than your own pages is set out in whether ChatGPT trusts press coverage more than your website.

#Which sources do AI assistants compare?

Almost anything public that names your firm. For a UK B2B firm, these matter most:

Most firms of your size do not have, and should not try to create, a Wikipedia article. Wikipedia has strict rules on notability and on people writing about their own organisations, and a company-written article is likely to be challenged or removed.

Wikidata is different. It is the open database of facts behind Wikipedia, anyone can edit it, and its data is released into the public domain, so search engines and AI developers are free to reuse it. A firm does not need a Wikipedia article to have a Wikidata item, but the item needs references to public sources, and other editors will check it. Our guide to fixing wrong facts lists who can edit each of these sources and the order to update them in.

#How do assistants settle a disagreement?

They weigh sources rather than count them. None publishes its rules, so what follows is what we observe in the answers we record, not a documented method.

Independent sources often beat your own. If your website says one thing and a respected trade publication says another, assistants often go with the publication, since you have a reason to flatter yourself and it does not. That cuts both ways: an old article that describes you wrongly can outweigh your accurate website.

Newer beats older when it explains the change. A source that says "Example Ltd, which moved its head office to Manchester last year" lets an assistant reconcile old and new. Without one, it sees a contradiction and hedges.

Specific beats vague. "Founded in 2009 by two former tax managers, with offices in Leeds and York" gets repeated. "An established practice serving Yorkshire" gets softened or ignored.

#What does this look like for a real firm?

TCC Group provides regulatory compliance for UK financial services through three brands: TCC for advisory and managed services, Recordsure for compliance AI and Momenta for specialist resourcing. Its case study puts the problem in one line: "Three brands, one group. AI could only see the pieces."

"ChatGPT already placed TCC top for outsourced suitability reviews. But on Consumer Duty and Section 166, the two areas the group backs hardest, the assistants named other firms and not TCC. One had stopped recognising the UK firm altogether, and none of them connected the three brands."

The evidence came from "36 recorded answers from ChatGPT, Gemini and Claude, every conversation kept and reproducible." And the cause was structural. "It was not a content problem." "What failed was underneath: a site organised around internal teams rather than the questions buyers ask, pages that half-answered the same question so neither was cited, and machine-readable data that declared a name, a logo and nothing else."

That is the name-consistency problem at group scale. The plan answered it with "Ten solution pages, one for each problem a buyer types, from Section 166 and FCA intervention to vulnerable customers, plus sector hubs and one group page telling the story once." "TCC chose to build it with its own team, and the new tcc.group is live, built to the plan, with nothing about it depending on Fortitude."

If your firm has merged, rebranded or runs several brands, the lesson carries over: publish one page that names every brand and how they relate, declare the same relationship in your structured data, and make each brand's profiles point to it.

#Why does ChatGPT give the wrong number of employees for my company?

Because headcount is one of the facts most often left out of date. In our experience assistants piece it together from your LinkedIn company page, business data sites, old press coverage and your website, and those sources rarely agree.

State a headcount range on your about page with the month, for example "about 90 people, September 2026", set the company size band on your LinkedIn page to match, and check what the business data sites show. Then the assistant has one current, dated figure to repeat instead of three old ones to reconcile.

#Does Companies House data affect what AI assistants say about my business?

Yes, mostly indirectly. Companies House holds your registered name, registered office, directors and nature of business, and business data sites copy it, so an old registered office or a departed director can echo across dozens of pages that assistants read.

Keep your filings current, and if your trading name differs from your registered name, say so on your website so an assistant can connect the two. A departed director still listed on data sites is a common source of confusion about who leads the firm.

#How do I update my business details in Bing so Copilot gets them right?

Claim your listing in Bing Places for Business and match it to your fact sheet, then check that Bing has indexed your key pages, using Bing Webmaster Tools if it has not. Copilot draws on Bing, and ChatGPT search has too.

This matters more than its share of web traffic suggests, because Copilot is what many buyers have at work. Microsoft 365 Copilot passed 30 million paid seats in July 2026, according to Microsoft's results, and Forrester found in January 2026 that business buyers are four times as likely as consumers to use Copilot. Check Apple Business Connect while you are there, since Apple Maps draws on it.

#How to tell whether your firm has a confidence problem

You can run this test in under an hour, and it is worth repeating every month.

#Step 1: Ask the assistants about your firm directly

Ask ChatGPT, Gemini, Claude, Copilot and Perplexity what your firm does, who its typical clients are and whether it is a good choice for your main service. Use a fresh chat each time and save each answer with the date.

#Step 2: Ask the question your buyers ask

Ask the category question, such as "Which firms should I consider for this service in this region?", and note whether you appear, where, and how you are described compared with competitors.

#Step 3: Mark the hedges and errors

Highlight every "appears to", every "according to its website", every generic description, wrong fact and omission. These are your confidence gaps.

#Step 4: Look at the sources the assistant used

Perplexity, Copilot and ChatGPT when it searches show the pages they drew on. Note which sources describe your competitors and whether any mention you. That is the start of your target list.

#Step 5: Record it and repeat

Keep a fixed list of 20 to 50 buyer questions in a spreadsheet and ask each several times a month in a clean session, because answers vary from run to run. Count a change only when it repeats. Our guide to finding out whether ChatGPT recommends your firm sets this up in more detail.

#How do we fix a confidence problem?

Make the sources agree, then give the assistant independent evidence that repeats what they say.

Settle one description and use it everywhere. One plain sentence saying what you do, who for and where, used on your website, LinkedIn, listings, directory entries, the boilerplate you give journalists and your speaker biographies. The independent coverage you earn will then tend to repeat it.

Complete the listings and databases. Google Business Profile, Bing Places, Apple Business Connect, Companies House, the business data sites and, where you can reference it, Wikidata.

Label the facts on your site. Structured data states your name, former names, address and official profiles, and for a group how its brands relate, in a form software reads directly. It helps search engines read your facts correctly; Google says its AI features need no special markup, and OpenAI and Anthropic have not said they read it, so keep it accurate and do not treat it as a ranking trick. Our guide to schema markup for AI assistants explains what to include.

Earn independent evidence. Most founder-led firms do it through a few repeatable routes: expert comment for the trade titles buyers read, original data about your market, guest articles an editor accepts on merit, case studies on clients' own sites, and roles with professional bodies. Which websites that mention you count with AI and digital or traditional PR for being named by ChatGPT cover the trade-offs. Avoid anything that promises links in bulk: it adds to the count without adding to the evidence.

#What to do this month

1. Run the five-step test above on your firm and your main category question, and save every answer with the date.
2. Write one plain sentence saying what you do, who for and where, and update your website, LinkedIn page and listings to match.
3. Claim or correct your Bing Places, Google Business Profile and Apple Business Connect listings, and check your Companies House details.
4. Ask the publishers of the three most visible out-of-date descriptions of your firm to correct them.
5. Choose one route to independent coverage, such as expert comment for your main trade title, and commit to it for three months.

#Frequently asked questions

#How do I change what ChatGPT says about my company?

Change the sources it reads. Telling ChatGPT it is wrong in your own chat may change that conversation, but not what other people see. Make your website, listings and profiles agree, correct third-party sources, and earn independent coverage that states the same facts. You can also flag a wrong answer with the thumbs-down button.

#Should my company have a Wikidata entry, and how do I create one?

Often yes, if you can support it with public references such as Companies House and independent coverage. Create an account, search to check no item exists, then add your firm with a neutral description, what kind of organisation it is, country, founding date, official website and headquarters, each backed by a reference. Write it like a fact sheet, not a brochure, because other editors will check it.

#Do mentions without a link help at all?

Yes, in our observation. Assistants read text, so a trade article that names your firm and describes what you do adds to the evidence even without a link. A link adds a little more, because it helps search engines connect the mention to your site.

#Can a critical article make assistants less likely to recommend us?

It can colour the answer, especially if little else has been written about you. Fix the underlying issue if the criticism is fair, ask for a correction if it is not, and build enough balanced coverage that one article is not the main thing an assistant finds.

#We rebranded. How do we stop ChatGPT using our old name?

You may not stop it entirely, and you do not need to. Aim for assistants to connect the two names: say plainly on your website that you were formerly known by the old name, list it as an alternative name in your structured data, and update every profile. A trade press announcement gives assistants an independent source that links both names.

#How long before new coverage changes what assistants say?

Assistants that search the web live (ChatGPT search, Perplexity, Copilot, Google's AI features) can reflect a new article within weeks. What a model knows from its own memory changes only when it is retrained, which takes months. In our experience, steady coverage over several months shifts answers more reliably than one big article.

If the answers you get back are vague or missing, Reputation That Wins is built to change them, starting with the sources the assistants already read about your competitors. To see where you stand first, book a free Growth Review: 30 minutes with David Adams and three fixes in writing.

#Sources