Blog

How to Monitor Your Brand in ChatGPT

Key Takeaways

  • Monitor your brand in ChatGPT by running the same 10 to 20 buyer questions on a schedule and logging each answer, or get a free AI search audit that shows where ChatGPT mentions you and which competitors it names instead.
  • Score each answer for four separate outcomes: a mention, a recommendation, an accurate description and a citation to your page or a third-party page.
  • Keep unbranded buyer questions apart from prompts that name your brand, because a branded prompt only tests whether ChatGPT describes you accurately.
  • Run each priority prompt signed out and signed in, each with and without Search selected, and record any search ChatGPT runs on its own.
  • Recheck high-value prompts weekly and the full set monthly, then update or publish the page that each recurring gap points to.

To monitor your brand in ChatGPT, run a fixed set of 10 to 20 buyer questions on a schedule, save every answer, and score mentions, recommendations, descriptions and citations separately. Run each prompt signed in and signed out, with and without Search selected, because the answers change with those conditions. One answer is a starting point, not a brand report. To see where ChatGPT mentions you today and which competitors it names instead, start with a free AI visibility audit.

You can run a useful audit with a spreadsheet and 10 to 20 prompts. Keep the wording fixed, save what ChatGPT says, and separate results by test condition.

Decide what counts as visibility

Record four outcomes for each answer:

  • Mention: ChatGPT names your brand, even in passing.
  • Recommendation: ChatGPT presents your brand as a fit for the buyer’s stated need. Record this separately from a passing mention.
  • Description: ChatGPT gets your category, customers, capabilities, and constraints right. Mark specific errors.
  • Citation: A visible source link points to your page or to a third-party page discussing your brand. Record those as owned and third-party citations separately.

Open every relevant citation. Check whether the page supports the nearby claim and whether its information is current. OpenAI warns that ChatGPT Search results and citations can be incorrect or incomplete, so a link alone does not establish that the answer is accurate.

Choose questions your customers might ask

Start with 10 to 20 prompts. That is a manageable scope for a manual audit, not a platform requirement. Pull questions from customer calls, sales objections, support tickets, and the jobs your product handles. A keyword export can help you find ideas, but it should not stand in for buyer questions.

Use a mix like this. The examples are illustrative prompts, not claims about how often buyers ask them.

Prompt group Example prompt What to inspect
Unbranded category “Which content marketing tools help a small B2B SaaS team publish regularly?” Whether ChatGPT names you without a hint.
Constrained use case “How can a two-person marketing team publish useful comparison pages every week?” Whether ChatGPT connects vendors to the team’s actual constraint.
Buyer shortlist “What should a B2B marketer compare before choosing an AI content tool?” Which vendors and selection criteria appear.
Competitor alternatives “What are alternatives to [competitor] for publishing B2B content with a small team?” Whether you appear on an existing shortlist.
Branded accuracy “What does [brand] do, and who is it for?” Whether ChatGPT repeats wrong or outdated facts.
Direct comparison “Compare [brand] and [competitor] for a two-person B2B marketing team.” Whether it describes meaningful differences accurately.

Keep branded accuracy prompts apart from unbranded discovery prompts. If you put your name in the question, a mention tells you little about whether ChatGPT would find you on its own.

Run four checks for each priority prompt

ChatGPT can search the web without you selecting Search. It can also use Search when you select it. Signed-out users can access web search, while signed-in search-query rewriting may use saved memories when enabled. Record what you selected and what you observed instead of calling an ordinary chat a guaranteed “no-search” test.

Test condition How to run it What to record
Signed out, ordinary chat Start a fresh signed-out chat and submit the prompt without selecting Search. Answer, citations, and whether ChatGPT visibly searched anyway.
Signed out, Search selected Start another fresh chat, select Search, and submit the exact same prompt. Answer, cited pages, brands, and recommendation language.
Signed in, ordinary chat Start a fresh chat outside a project or established thread. Do not select Search. Account, personalization settings, answer, citations, and visible search activity.
Signed in, Search selected Start another fresh chat, select Search, and repeat the prompt. Sources, brands, descriptions, and differences from the signed-out run.

Label search behavior Search selected, search observed automatically, or no search observed. The last label describes what you could see. It makes no claim about hidden system behavior.

Keep your country and language in the log. ChatGPT can use approximate IP-based location for search, and optional device location can add detail. Projects can add instructions, files, and prior-chat context, so keep project chats out of your baseline.

If you use Temporary Chat for a less personalized signed-in check, inspect the options in your account. OpenAI describes an Unpersonalized option that excludes memories, custom instructions, and plugins. Record the option you selected rather than treating every Temporary Chat as unpersonalized.

Save each answer and score it

Use one spreadsheet row per prompt, run, and session condition. These columns give you enough detail to compare answers later:

prompt ID | exact prompt | intent group | date and time | country | language | signed-in status | account or workspace | personalization setting | model if visible | Search selected or observed | brand mentioned | recommended or merely named | first brand named | competitors | description accurate | owned citation | third-party citation | cited page title | screenshot or answer copy | action

Save the full answer or a screenshot. For citations, record the page title and URL, then open the page. A source badge without the underlying page gives you little to act on.

Report rates only for a defined group and condition. If your brand appears in 6 of 20 valid unbranded, signed-out, Search-selected answers, its observed mention rate for that test set is 30%. The denominator is those 20 answers, not every conversation your customers have with ChatGPT. Report branded prompts separately.

You can also count recommendations and accurate descriptions within that same group. A company that appears often for the wrong buyer, or with the wrong capabilities, needs a different fix from a company ChatGPT never names.

Recheck weekly, then review monthly

Check a few high-value prompts each week and rerun the full set monthly. Keep prompt wording and session conditions consistent. Run an extra check after a major product change, a correction to an inaccurate page, or publication of a page meant to answer a tracked question.

Repeated checks matter. In Glen Allsopp’s 28-day study of more than 70,000 ChatGPT and AI Overview responses, 41% of the ChatGPT prompts never returned the same set of brands twice across the month. The study used primarily commercial prompts adapted from search terms, so its figures describe that sample rather than every B2B category.

Compare repeated answers before assigning work. One omission may be normal variation. Several omissions under the same recorded condition give you a better reason to inspect your pages and the sources ChatGPT used.

Decide what to publish or fix

Start with the exact buyer question and the pages ChatGPT cited. Then choose the smallest useful piece of work.

Recurring observation What to check What to do
You are absent from unbranded category prompts. Read your category and use-case pages beside the pages ChatGPT cites. Clarify whom you serve and what you do on an existing page, or publish a useful page if that answer is missing.
ChatGPT names you but describes you incorrectly. Find the owned or third-party page behind the claim. Update product, pricing, integration, or documentation facts, and request corrections to outdated third-party claims where possible.
ChatGPT recommends a competitor for a narrow use case. Check the buyer’s criteria and whether your product actually meets them. Publish a substantive use-case page or fair comparison only when you can support the case with facts.
ChatGPT mentions you but rarely cites your pages. Check whether an owned page answers the question directly and is accessible. Improve the relevant page and check access. Do not assume schema markup will secure a citation.
An owned page is absent from Search answers. Ask your technical owner to inspect robots.txt, CDN rules, firewall rules, and crawler access. Check access for OAI-SearchBot, then continue testing.

OpenAI identifies OAI-SearchBot as its crawler for search inclusion. GPTBot controls a separate potential training use, so you do not have to allow training to make a public page eligible for search. If you inspect server logs, keep ChatGPT-User separate too: it can visit pages for certain user-initiated actions, but it does not control automatic search inclusion. Allowing OAI-SearchBot makes inclusion possible, not guaranteed.

When the audit points to a genuine content gap, Evatype can help your team research buyer questions and competitors, draft and review on-brand pages, and publish them directly to your website. Use your prompt log to decide which question deserves a page first.

Check clicks separately from answers

Your prompt log records what appeared in the answers you tested. Your analytics record something else: visits from people who clicked through. OpenAI says ChatGPT Search referral links automatically include utm_source=chatgpt.com, which you can look for in analytics. That report cannot count someone who saw your name in ChatGPT and reached you another way.

Keep both measures. They answer different questions, and neither represents every buyer’s ChatGPT session.

Know when a tracking tool earns its cost

A spreadsheet works for a small prompt set. If you need repeated runs across many clients, markets, languages, or prompt groups, a tracker may save time.

Before buying, ask which ChatGPT experience it tests, whether it selects Search, how it handles signed-in status and personalization, which locations it uses, and whether you can inspect raw answers and citation URLs. Make sure you can export the evidence. A tool’s results still describe its tested sessions and settings, not every answer ChatGPT shows.

Frequently Asked Questions (FAQ)

Is my brand mentioned in ChatGPT if I ask for it by name?

A branded prompt tells you whether ChatGPT can describe your company. To check discovery, ask buyer questions that do not include your brand name and track those results separately.

Why does ChatGPT mention us when I am logged in but omit us when I am logged out?

Check the account’s memory and instructions, location, visible search activity, and any project context before assigning a cause. Repeat the exact prompt in fresh chats and keep the two session conditions separate.

Can I test a ChatGPT answer without web search?

You can submit a prompt without selecting Search and record whether you observe search activity. ChatGPT may search automatically, so label the run “no search observed” rather than claiming that no search occurred.

How often should I check ChatGPT brand visibility?

Check a small set of high-value prompts weekly and review the full set monthly. Rerun relevant prompts after a major product change or a page update intended to answer them.

Will allowing OAI-SearchBot make ChatGPT cite our pages?

Allowing OAI-SearchBot helps make public pages eligible for ChatGPT Search inclusion, but OpenAI does not guarantee placement or citations. Check access first, then judge results through repeated answers.