Why does ChatGPT describe my firm wrongly, and how do I fix it?

Ask ChatGPT what your firm does and you may get an answer that is two years out of date and slightly wrong. It names a service you stopped offering, gets the founding year wrong, and calls you "a marketing agency" when you have spent three years repositioning as a specialist consultancy. Ask Gemini and you get a different wrong answer. Neither is lying. Both are doing their best with what the web says about you, and the web disagrees with itself.

That first description matters more each year. 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. G2 surveyed software buyers, so the figure will differ for services, but for many buyers the assistant's summary is now the first account of your firm they read.

In our experience it is one of the most common problems founder-led firms have with AI search, and one of the most fixable. AI assistants try to understand your firm as a single thing, an entity in their terms, and describe it in a sentence. When their sources contradict each other, they pick the wrong version, repeat an old one or leave you out.

#Why do AI assistants get facts about my firm wrong?

Because they compare many sources, and those sources disagree. Your website, LinkedIn, Companies House, directories, reviews and old press each hold a slightly different version of what you do, where and since when, and the assistant has to choose.

That causes four problems. Misattribution: an assistant may attach a claim to the wrong version of your firm, or describe a dropped service as current. Lost confidence: it qualifies what it says, a problem covered in why ChatGPT sounds unsure about your firm. Exclusion: asked for the best firms in a category, it may leave out one whose category is unclear, in our reading of the answers we record. Exposure: a negative review has no clear story to set against it.

#Which sources do AI assistants compare?

Almost anything public that names your firm: your website, Companies House, LinkedIn, your Google, Bing and Apple listings, Wikidata, Crunchbase, business data sites, directories, press coverage, client case studies and speaker biographies. The table in Step 1 lists who can edit each.

Bing matters more than most firms assume, because Copilot draws on it and ChatGPT search has too. How assistants weigh your own sources against independent ones is covered in how AI cross-checks what the web says about you; this guide is about correcting the wrong facts.

#Why do ChatGPT and Gemini give different answers about my firm?

Because they read different sources and remember different snapshots of the web. Gemini and AI Overviews start from Google's index and Copilot from Bing's; in a small Seer Interactive test published in February 2025, 87% of ChatGPT search citations matched Bing's top results. A firm with a stale Bing listing can be described correctly by Gemini and wrongly by Copilot on the same day.

Each assistant also has favourite sources. Muck Rack's May 2026 analysis of more than 25 million cited links found ChatGPT's most cited domain was Wikipedia and Gemini's was Reddit. One model's memory may also predate your rebrand while another's does not, and answers vary from run to run, so ask each question several times in a fresh chat and note what repeats. Our guide to which AI assistants buyers use goes further.

#How do assistants decide which version to believe?

No assistant publishes its rules, so what follows is what we observe in the answers we record. Assistants compare meaning, not words: "employment law for manufacturers" and "HR consultancy" read as two different firms. They cross-check claims: four offices on your site and two on LinkedIn may become a range, or nothing at all.

They also seem to favour your website over LinkedIn, LinkedIn over directories, and recent statements over old ones. The pattern breaks easily: a detailed recent trade article can beat a thin, stale website. It is plausible that a neutral write-up saying you are slow to respond could outweigh a homepage promising a one-hour response; we have not tested that directly, so treat it as a risk to manage.

#Which errors matter most?

In our experience the priorities fall into three tiers, and positioning comes first because recommendations depend on category.

Fix minor conflicts too, because they suggest a firm that is not paying attention.

#Why does ChatGPT say we offer a service we stopped years ago?

Usually because something public still says you do, or the model learned it before you stopped: an old service page still live or indexed, a directory profile written years ago, an old press release, or a training snapshot that predates the change.

Fix what is live first. Redirect or rewrite the old page, update every directory profile, and add one plain sentence to your about page, for example: "We stopped offering payroll in 2023 to focus on advisory work." Assistants that search the web then find a current source that explains the change; answers from memory catch up when the model is retrained.

#How to fix wrong facts: the master fact sheet and four-week plan

One organised person can do most of this in four weeks.

#Step 1: Build a master fact sheet

Make a spreadsheet. Down the left: what you do in one sentence, who you serve, sectors, locations, founding date, leadership, size, accreditations and former names. Across the top, one column per public source. Fill in what each says today and the conflicts appear at once. Mark the facts that must mean the same thing everywhere: what you do, for whom, where, since when and who leads the firm. These are the columns most firms need:

#Step 2: Trace each conflict to its cause

For example, the website says founded 2015 because that was the rebrand, LinkedIn says 2014 because that was incorporation, and a directory copied the website years ago. Knowing the cause tells you what to change.

#Step 3: Decide the truth and write it on one page

Choose one version of each fact and commit. Write it on one page, with your one-sentence description, three differentiators and former names, and check every profile and biography against it from now on.

#Step 4: Update every source in a fixed order

Week one: website and LinkedIn company page. Week two: Google Business Profile, Bing Places, Apple Business Connect, Companies House and the knowledge panel. Week three: directories, review sites, business data sites, Crunchbase, Wikidata and staff LinkedIn profiles. Week four: email signatures, proposals and pitch decks, so nobody reintroduces the old version.

#Step 5: Explain changes rather than hide them

If you have repositioned, say so on your site: "We began in 2015 as a general marketing agency and since 2023 have focused on demand generation for industrial firms." That explains the conflict an assistant will still find in old coverage.

#Step 6: Publish the update and label the facts

A short LinkedIn post and a dated page on your site give assistants that search the web a current source. Structured data, code that states your name, former names, address and official profiles for software to read, helps search engines read those 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; see schema markup for AI assistants.

#How do I update my company's Google knowledge panel?

You claim it, then suggest edits. Google says knowledge panel information "generates automatically based on public information on the web", so you cannot write it, but you can ask for changes.

Search Google for your firm. If the panel shows "Claim this knowledge panel", click it and sign in to one of the official sites or profiles Google lists, such as YouTube or Search Console, as set out in Google's verification guide. Once verified, search for your firm while signed in, click "Suggest edits", choose the wrong item, describe the correction and add supporting links. Google says it reviews feedback from verified users within a few days, but may not make a change it cannot confirm. Not every panel can be claimed yet.

#How do I report a wrong answer directly?

Use each assistant's feedback controls and, for false statements about a named person, the data protection route. Do both alongside fixing the sources.

Data protection law (UK GDPR) protects information about people. If ChatGPT says something false about you or a co-founder by name, OpenAI's Europe privacy policy, which covers UK users and was updated in August 2026, says you can request a correction or removal, and that OpenAI "will consider your request based on applicable law and the technical capabilities of our models". The ICO says an organisation normally has one month to respond to a correction request, and that if you are still unhappy after complaining to it, you can complain to the ICO. Wrong facts about the company itself are not personal data, so fix the sources.

#Described correctly, still not recommended

Fixing wrong facts makes you describable. It does not, on its own, get you recommended, and ID-Pal shows the gap. ID-Pal provides identity verification for fund administrators, wealth managers, law firms and accountants, and had bought NorthRow, a UK business verification platform, and folded it into id-pal.com.

The assistants got the basics right: "6 of 6 correct answers to 'what does ID-Pal do?', every engine, every run." But "AI could describe ID-Pal. It never recommended it." There was a conflict too, of exactly the kind this guide describes: "The redirect from northrow.com was technically correct, but nothing on either site told Google or the AI engines that NorthRow and ID-Pal were the same company."

The fix had two parts. For the conflict: "About four hours of structured data work declaring NorthRow part of ID-Pal on both domains. Live in week one, with no copy changes and no risk to the redirect." For recommendation: "Five industry pages rebuilt: fund administration, wealth management, legal, accountancy and estate agency. Each opens with the question a compliance lead types, names the regulator, and answers inside the first ninety words."

If you are described correctly and still not named, read why ranking first on Google does not guarantee a mention.

#How do we keep it consistent?

Give one person ownership and check against the fact sheet every quarter. Conflicts come back when marketing updates the website, sales updates LinkedIn and nobody updates the directories. Define the triggers that force an update everywhere at once: new leadership, positioning or offices, a new accreditation, a dropped service. The trust signals assistants look for are set out in what E-E-A-T is and why AI cares about it.

#What to do this month

1. Build the master fact sheet: ten facts down the side, every source in the table above across the top.
2. Ask ChatGPT, Gemini, Claude, Copilot and Perplexity "what does [your firm] do?" three times each in a fresh chat, and log every error.
3. Decide the truth for each conflict and write the one-page source of truth.
4. Update your website and LinkedIn this week, claim Google Business Profile, Bing Places and Apple Business Connect next week, then work through the rest in order.
5. Report remaining errors with each assistant's feedback button, and claim your knowledge panel if Google offers it.
6. Put a quarterly check in the diary with a named owner.

#Frequently asked questions

#How do I correct wrong information ChatGPT says about my company?

Fix the sources it reads, then report the answer with the thumbs-down button. Correct your website, LinkedIn and listings against a master fact sheet and ask directories and publications to correct theirs. There is no form that edits what ChatGPT says about a company.

#Can I complain to OpenAI if ChatGPT says something false about my business?

You can report it with the thumbs-down button. If the false statement is about a named person, such as a founder, UK data protection law gives that person the right to ask for a correction through privacy.openai.com or dsar@openai.com, and the ICO handles complaints about the response. Facts about the company itself are not personal data, so fix the sources. If a statement is seriously damaging, take legal advice.

#How long until the assistants describe us correctly?

Assistants that search the web live (ChatGPT search, Perplexity, Copilot, Google's AI features) can reflect corrections within weeks. Answers from a model's own memory change only when it is retrained, which takes months. ID-Pal is a dated example. AI answers still recommending NorthRow as a rival: "six in May, one in September 2026." In our experience descriptions usually improve within a quarter of fixing the serious conflicts, but nobody can guarantee what an assistant will say.

#Will conflicting information hurt our Google rankings?

Probably not much in the classic results, which rank pages. Google's AI Overviews and AI Mode describe your firm as a whole, so inconsistency can cost you there. Ranking helps you get cited but does not guarantee it, as we explain in whether you still need SEO now buyers ask ChatGPT.

#Should we delete old information or update it?

Update where you can. Deleting leaves a gap that older copies elsewhere fill with the wrong version. Old, accurate history is fine; old, contradictory positioning is not.

#A trade publication described us wrongly. Can we fix that?

Ask for a correction; most publications will fix factual errors. One wrong article against consistent official sources is a minor problem; one that matches an old version of your own site is a serious one, so make your own site the stronger source. Our technical checklist for an AI-ready website covers what else it needs.

Assistants can only describe the firm the web shows them, so give them one firm to describe. To see how you are described today, book a free Growth Review: 30 minutes with David Adams and three fixes in writing. Making press, listings and reviews tell one story is the work of Reputation That Wins.

#Sources