ChatGPT is often the first — sometimes the only — place someone asks about your services, your programs, or your organization by name. This guide walks through a one-hour check of what it actually says: how to test it, what to record, and how to trace wrong or missing answers back to pages you can fix.
By Matt·Updated July 12, 2026
Quick answer
How do you check your organization in ChatGPT?
Open a fresh or temporary ChatGPT chat and ask the real questions your audience asks — about your topic, your services, and your organization by name. Run each question twice: once letting ChatGPT answer from memory, once telling it to search the web. Grade every answer for whether you appear, whether the facts are current, and which sites it cites. Errors almost always trace back to your own pages or a stale third-party listing.
ChatGPT builds answers from its training data plus live web search, and when it searches it leans heavily on established reference sites and directories — as of July 2026, how third parties describe you matters as much as your own website. (Background: how ChatGPT finds information.) This page covers one engine; the full AI visibility audit covers them all.
Why check ChatGPT specifically
When someone asks ChatGPT "which accounting firms near me work with nonprofits?" or "is the community health clinic on Fifth Street taking new patients?", they read one answer and act on it. If that answer names a peer organization instead of you, or repeats an address you left two years ago, you'll never see it happen — the conversation leaves no trace in your analytics unless the person clicks a cited link. (When they do click, ChatGPT tags the link so the visit is identifiable — see AI referral traffic.)
ChatGPT also differs from the other engines in a way that changes what you audit: when it searches the web, it favors established directories, reference sites, and news coverage over small organizations' own websites. Your site being accurate is necessary but not always sufficient — the third-party pages that describe you carry unusual weight here.
Three ChatGPT behaviors this check accounts for
Two sources, two failure modes. ChatGPT answers either from its training data (a memory that can be years stale) or from a live web search. Each can be wrong for a different reason, so the audit tests both.
Third parties speak for you. Directory listings, chamber-of-commerce pages, old news stories, and reference sites often shape ChatGPT's answer more than your homepage does.
Answers vary run to run. The same question can produce different answers and sources each time. Look for patterns across your whole question list, not a verdict from one answer.
How to run the check, step by step
Five steps, about an hour, ending in a short list of pages to fix.
1
Write down your facts before you ask anything
List your organization's correct details on one sheet: exact name, locations, hours, current programs or services, who you serve, and current leadership. This is your answer key — without it, a confident-sounding ChatGPT answer that's 80% right will slip past you. A state professional association, for example, should include its current executive director and this year's conference dates, because those are exactly the details that go stale.
2
Open a clean session
Use a temporary chat (or log out entirely) so your own history and saved preferences don't color the results. You're trying to see what a stranger sees, and ChatGPT personalizes answers for signed-in users who've chatted with it before. A free account is fine — as of July 2026 the free tier includes the web search this check depends on.
3
Test its memory first, then its search
Ask each question plainly first and note whether ChatGPT answers without searching — that's what it "remembers" about you from training data, and it's often years old. Then ask again with "search the web for…" at the front. Comparing the two tells you where the problem lives: a wrong memory answer but a correct search answer means old information is still circulating; wrong both times means the live web itself — including missing or inconsistent information — needs fixing.
4
Run branded and unbranded questions
Branded questions ("what does [organization] do?") test accuracy. Unbranded questions ("where can families get free tax help in [county]?") test whether you're recommended at all — usually the more sobering result. Use the prompts below as starting points, and record each answer the same way: mentioned or not, accurate or not, cited or not, and who else was named. The AI visibility scorecard gives you a ready-made grid for this.
5
Follow every citation to its source
This is the ChatGPT-specific payoff. Click each link it cites and note the site. If it's describing you from a stale chamber listing, an old news story, or a directory you forgot existed, those pages are your fix list — and they're often fixable with an email. If it cites a peer organization's pages instead of yours for questions you should own, analyze what their footprint has that yours doesn't.
Diagnostic prompts to run
Copy, paste, and fill in the brackets. Run each in a fresh chat.
Baseline description check (branded)
Prompt
What can you tell me about [organization name] in [city, state]?
Include what it does, who it serves, its current programs or
services, and where you found this information.
Recommendation test (unbranded)
Prompt
I'm looking for [the service you provide] in [county or city].
Which organizations would you recommend, and why? Please search
the web and cite your sources.
Freshness check (forced search)
Prompt
Search the web for the current [hours / schedule / leadership /
programs] of [organization name]. List every page you used, and
flag anything that looks outdated or contradictory.
What the responses tell you
Four common patterns, and what each one means.
Accurate, and your site is cited. The healthy result. Note which of your pages earned the citation — that tells you what ChatGPT considers your most quotable content, and it's a model for the rest of your site.
Mentioned, but never linked. ChatGPT knows you exist — usually from third-party pages — but isn't treating your site as a usable source. That points to pages that are hard to read (information locked in PDFs, images, or script-driven widgets) or that never state the basic facts in plain text. The website readiness check will find the specific cause.
Confidently wrong. Former staff, discontinued programs, an old address. The freshness prompt above will usually expose the stale source. Correcting it — a directory listing, a partner page, your own forgotten page — is the highest-urgency fix in this whole audit.
Absent entirely. On unbranded questions you should own, ChatGPT recommends others. Because ChatGPT leans on third-party coverage, the fix is usually off-site: current listings, local news mentions, partner links. This is also where ChatGPT differs most from its siblings — Claude responds more to your own site's plain factual prose, and Perplexity to fresh, crawlable mentions anywhere.
What this check turns up in practice
Two representative scenarios — one absence problem, one accuracy problem.
The association losing its own members' questions
What the check found
A state professional association asks ChatGPT "where can [professionals] in [state] get continuing-education credits?" and gets a list of national online vendors — never the association, which runs the state's largest CE program. Following the citations shows why: ChatGPT sourced the answer from two national directories, and neither listed the association. Its own CE pages described "member growth opportunities" without ever saying "continuing-education credits."
What it changed
The association rewrote its CE page to state plainly what credits it offers, for whom, and at what cost, and got itself added to the two directories ChatGPT kept citing. On rechecks a few weeks later, it appears in the recommendations — cited from its own site.
The lesson: ChatGPT's answer was built entirely from third-party pages. The association didn't need better marketing — it needed to exist, in plain words, on the pages ChatGPT was already reading.
The accounting firm with a retired partner
What the check found
A 20-person accounting firm runs the branded prompt and gets a mostly flattering answer — led by a named partner who retired three years ago and the firm's previous office address. The memory-versus-search comparison in step 3 shows the wrong details in both modes: an old chamber-of-commerce listing and a 2021 news story were feeding the live search, and the firm's own site buried its current team in a PDF brochure.
What it changed
The firm published a plain-HTML team page with current bios, updated the chamber listing with one email, and put the current address in its site footer. Rechecks show ChatGPT naming the right managing partner and citing the firm's own team page.
The lesson: a prospective client had no way to know that answer was stale. Accuracy errors surface fastest with branded prompts run in both memory and search modes — and the fixes here took an afternoon.
Why does ChatGPT link to other organizations but only mention ours in plain text?
A link means ChatGPT pulled that organization's page in during a live web search and used it as a source. A plain-text mention usually means it knows about you only from older training data or from third-party sites. The fix is making your key pages easy to find and read — clear factual text, current details, and consistent listings on the directories and reference sites ChatGPT tends to cite.
Does building a Custom GPT for our organization improve how ChatGPT answers about us?
No. A Custom GPT is a private, separate assistant that only helps people who open it deliberately. It has no effect on what the main ChatGPT says when someone asks about your topic or your organization. Improving those answers means improving your public website and the third-party pages that describe you.
ChatGPT said something wrong about us. Can we get it corrected?
There is no form to submit a correction to. ChatGPT's answers come from its training data and from live web pages, so the practical fix is to correct the sources: update your own site, then fix the stale directory listings, old news mentions, and outdated profiles it is drawing from. Answers built from live web search typically reflect those corrections within weeks; the older training-data memory takes longer to catch up.
Do we need a paid ChatGPT account to run this check?
No. As of July 2026 the free tier includes web search, which is the capability this audit depends on. A free account and a temporary chat are all you need. Paid plans add features, but none of them change how ChatGPT describes your organization to the public.