By Mehdi HassanWhat does ChatGPT say about my company, and why does it matter?
ChatGPT says whatever the sources it reads, and the patterns it learned in training, suggest about your company, which may be accurate, outdated, vague or missing entirely. It matters because buyers now ask assistants to shortlist and compare vendors before visiting any website. Forrester found 94% of business buyers used generative AI in their last purchase. If ChatGPT describes you wrongly, or names three competitors and not you, that shapes the shortlist before you ever hear about the deal.
The good news is that the answer is usually traceable. When ChatGPT searches the web, it cites the pages it used. Those citations show you which page said what, and that tells you what to change. This how-to is part of our hub on getting your brand recommended by ChatGPT. It gives you an audit template, a way to diagnose each problem, and the fixes that actually move the answer.
How do you run a ChatGPT brand audit?
Run a ChatGPT brand audit by asking a fixed list of prompts in a clean session, twice each, and logging the same fields every time. The fixed list is the point: it lets you compare month to month instead of reacting to one odd answer.
Build a prompt list of 20–30 questions
Split it into four groups: brand prompts ("What is [company]?"), category prompts ("Best [category] for [buyer]"), comparison prompts ("[company] vs [competitor]") and fact prompts ("How much does [company] cost?", "Does [company] integrate with [tool]?"). Take the wording from sales calls and support tickets.
Use a clean session
Log out or use a fresh account with no memory or custom instructions, and make sure web search is on. Your own chat history can bias the answer toward you.
Run each prompt twice
Answers vary between runs. Two runs per prompt is a reasonable minimum for a monthly check.
Log five fields
Mentioned (yes or no), position in the answer, description used, factual errors, and cited URLs.
Repeat in other assistants
Run the category and comparison prompts in Perplexity, Gemini and Copilot as well. They use different indexes, so a gap in one isn't a gap in all.
Score and prioritize
Rank issues by commercial impact: a wrong price or a missing category recommendation matters more than an awkward phrasing.
Prompt type | Example | What it reveals |
|---|---|---|
Brand | What is [company] and who is it for? | Whether ChatGPT understands your positioning |
Category | Best [category] tools for [buyer type] | Whether you're recommended at all |
Comparison | [company] vs [competitor] | How you're framed against alternatives |
Fact | How much does [company] cost? | Pricing, plan and integration accuracy |
Problem | How do I solve [problem] without [workaround]? | Whether you're tied to the job you do |
How do you find out why ChatGPT has wrong information about your company?
Find the cause by reading the cited sources for each wrong answer and matching the error to a page. Most errors fall into a handful of patterns, and each one has a different owner.
Symptom | Likely cause | Where to fix it |
|---|---|---|
Wrong or old pricing | Pricing hidden, in an image, or outdated on a third-party site | Your pricing page, then review and directory listings |
Vague or wrong description | Homepage hero doesn't state category and buyer | Your homepage and About page |
Not mentioned for your category | Few third-party sources tie you to that category | Reviews, comparisons, community and partner pages |
Old product or old positioning | Stale profiles, press or training data | Third-party profiles, then patience for model updates |
Missing features or integrations | Features buried in JavaScript tabs or docs only | Feature and integration pages in plain HTML |
Confused with another company | Similar name, weak entity signals | Consistent name, Organization schema with sameAs, profile links |
If the answer has no citations, ChatGPT probably answered from training data. That kind of error is slower to change, because it depends on future model updates. Your best lever is still the same: make the current web consistent, so any search-grounded answer contradicts the old belief.
How do you fix what ChatGPT says using your own website?
Fix on-site causes by making your key facts explicit, in plain text, at the top of the pages assistants read. ChatGPT can only repeat what it can find and read. Vague hero copy, pricing in images and features hidden behind tabs all leave gaps that the model fills with guesses or with other people's descriptions.
- Homepage first screen: name the category, the buyer and the outcome in one sentence. "[Company] is a [category] for [buyer] that [outcome]."
- Pricing page: publish prices or the pricing model in an HTML table, with plan names and what each includes.
- About page: founding year, location, team, who you serve and what you don't do. These are the facts assistants use to describe a company.
- Comparison pages: one page per alternative buyers mention, with an honest statement of when each is the better choice.
- Trust and security page: for AI and data products, answer the data and privacy questions directly. See what to say on an AI trust page.
- Technical access: server-render these pages and allow OAI-SearchBot in robots.txt.
Schema markup won't fix a wrong answer on its own. Ahrefs' 2026 test found no citation lift from schema in ChatGPT. Organization markup with sameAs links does help search engines tie your profiles together, which is useful for name confusion. The detail is in does schema markup help you get cited.
How do you fix third-party sources ChatGPT relies on?
Fix third-party sources by updating the profiles you control, then earning new mentions on the sites ChatGPT already cites for your category. Your audit's cited URLs are the target list. If ChatGPT keeps citing the same review site, comparison article or community thread for a category prompt, that's where you need to appear.
Off-site work matters more than most teams expect. Ahrefs' study of 75,000 brands found web mentions correlated 0.664 with AI Overview visibility, against 0.218 for backlinks and 0.326 for Domain Rating. The pattern points to being discussed by name, in context, rather than collecting links.
- Update every profile you control: review platforms, directories, marketplace listings, LinkedIn, Crunchbase and partner pages. Use the same one-sentence description everywhere.
- Ask recent happy customers for reviews that name the use case and the buyer type, not just a rating.
- Contact authors of cited comparison articles with corrected facts, politely and with a source.
- Answer genuine questions in communities where buyers ask for recommendations, disclosing who you are.
- Publish something others will cite, such as original data or a free tool, and pitch it.
The full playbook is in brand mentions are the new backlinks.
For example, suppose ChatGPT says you're "a design tool for agencies"
For example, suppose a hypothetical B2B scheduling startup finds that ChatGPT describes it as "a design tool for agencies", a positioning it abandoned two years ago. The cited sources are an old directory listing, a 2023 launch article and the company's own homepage, whose hero says only "Work, beautifully."
The fix runs in order. Rewrite the homepage hero to say "[Company] is scheduling software for B2B sales teams." Update the directory listing and every profile with the same sentence. Publish a short changelog or About page note explaining the pivot. Ask five current customers for reviews that mention sales scheduling. Then re-run the audit monthly and watch which cited sources change first. This is an illustration of the process, not a real result.
How do you track whether the fixes change ChatGPT's answer?
Track the fixes by re-running the same prompt list monthly and comparing the five logged fields, then checking whether AI-referred visits and self-reported attribution move. Page fixes can appear within weeks once crawled. Training-data errors take longer.
- Prompt log: a spreadsheet with one row per prompt per month. Watch mention rate on category prompts and error count on fact prompts.
- GA4 channel group: group referrals from chatgpt.com, perplexity.ai, claude.ai, gemini and copilot. Our guide to tracking AI traffic in GA4 has the setup.
- Self-reported attribution: add "How did you hear about us?" to demo forms. Many AI-influenced buyers arrive later through search or direct. B2B attribution when buyers research in ChatGPT explains why.
- Tools: if your prompt list grows past what one person can run monthly, compare trackers in AI visibility tools compared.
