If you’re asking why ChatGPT recommends some brands and leaves yours out, start with the buyer’s question—not a supposed secret ranking formula. What was requested, what reasons were given, and what evidence can you inspect?
For a marketing leader, the next move is a diagnosis. Separate a brand mention from a useful recommendation, check the facts behind the answer, and choose work that makes your business easier for a buyer to evaluate.
How should you interpret a ChatGPT brand recommendation?
Treat a recommendation as an answer to a particular request, not a certificate that one business is best. Investigate the buyer’s needs, the answer’s stated reasons, and any linked evidence. Your next marketing decision should address a specific gap in buyer fit, public information, or factual accuracy—not merely your absence from a list.
What should you inspect before changing your marketing?
Save the exact prompt and the complete answer. Include the date, the product or model shown, whether search was used, and whether the conversation contained earlier instructions. Record those details rather than relying on a cropped screenshot.
Next, read the answer as a buyer would. Does it explain who each option serves? Does it discuss the constraints in the question? Does it name a brand without giving a meaningful reason to choose it?
Check every statement about your business against your current public information. A mention with the wrong service, location, or customer profile should go into the correction queue, not the success column.
Then inspect any linked sources. Read the relevant passage on the source page, not just its title. Note whether the source supports a company fact, a comparison, or the recommendation itself.
Do not treat the assistant’s explanation of its own selection as an audit of its internal decision process. Use the explanation to identify statements worth checking. Keep your conclusions tied to what the saved answer and source pages actually show.
For the current search experience and its controls, use OpenAI’s ChatGPT search documentation as an implementation reference when setting up your tests.
Are you measuring a mention, a citation, or a recommendation?
Use three separate labels in your audit. A mention means the answer names the brand. A citation means the answer displays a source link. A recommendation means the answer presents the brand as an option for the buyer’s request.
Those labels answer different business questions. If the goal is to enter a shortlist, a link to your educational article is not the same outcome as a recommendation for your service. If the goal is to correct misinformation, an accurate description may matter more than list position.
What does each observed result tell you to check?
| Observed result | Useful question | Next action |
|---|---|---|
| Your brand is recommended with a clear reason. | Is the reason accurate and relevant to this buyer? | Validate the facts and record the stated fit. |
| Your page is cited, but your brand is not recommended. | Was the page used to explain the topic or evaluate a provider? | Read the cited passage and classify its role. |
| Your brand appears with incorrect details. | Which public pages contain the correct information? | Check your owned pages and the sources displayed in the answer. |
| Your brand is absent. | Does this prompt describe a buyer you actually serve? | Check buyer fit before commissioning new content. |
Why does the buyer’s question matter so much?
Make buyer fit your first filter. Do not set a goal of appearing in every category answer. Choose the situations where your business has a clear, defensible reason to be considered.
For example, test a prompt such as: “Which payroll provider should a small business consider if it needs help with multistate hiring and has no in-house HR team?” This is an illustrative buyer scenario, not a recorded platform result.
That question gives your audit specific work to do. Inspect whether the answer addresses multistate support, the buyer’s staffing limits, and the kind of help each provider offers. Compare those requirements with what your own service pages actually explain.
Build the prompt set from real sales questions where you have access to them. Ask your team what buyers need to know before a first call, which objections recur, and what makes an opportunity a poor fit. Do not claim to have reviewed call recordings or CRM records that you have not accessed.
Separate branded and unbranded tests. “What does our company do?” is an accuracy check. “Which provider fits this problem?” is a consideration check. Keep their results apart so a strong branded description does not hide weak visibility in discovery questions.
What should a company improve first?
Start with the smallest useful correction. Do not order a site-wide rewrite because of one missing brand mention.
Make the offer specific enough to evaluate
Read your core service page without relying on the sales deck. A buyer should be able to identify what you provide, who it is for, where you operate, and what the engagement includes.
Replace broad claims such as “solutions for every business” with accurate scope. Explain meaningful limits too. If a service is designed for established teams rather than first-time founders, say so. The goal is a clearer buying decision, not a claim that wording controls an AI answer.
Keep the same core facts across channels. Use omnichannel marketing planning to assign ownership for website copy, public profiles, campaign messages, and sales materials. Give someone responsibility for resolving contradictions.
Answer the comparison questions buyers actually ask
Choose content around a decision: who the service fits, what it includes, what alternatives exist, and what a buyer should check before committing.
A useful comparison should state its criteria and treat alternatives fairly. Do not disguise a promotional list as independent research. If you discuss cost, publish only pricing or scope information you can stand behind.
Keep answers close to the relevant service. A separate article can explain a broad decision, while the service page should still provide the facts needed to evaluate the offer. Avoid publishing several thin pages that repeat the same answer under slightly different headings.
Use evidence that supports the actual claim
Publish a case study only when you can substantiate its details and have permission to use them. Define the starting point, the work performed, the measurement period, and the limits of the result.
Do not turn a client logo, image label, or testimonial teaser into proof of an outcome. If a particular result cannot be supported, leave that result out. Useful explanations do not need invented numbers.
Treat third-party coverage the same way. Pursue relevant editorial opportunities, but do not present paid placement or your own contributed copy as independent endorsement.
Assign technical checks to the right owner
Have the technical team check whether important pages are reachable, readable, and consistent with the intended indexing and access settings. Keep a record of any changes and the pages affected.
Require a specific implementation task rather than a vague promise to “make the site AI-ready.” When evaluating generative engine optimization services, ask which page, access issue, information gap, or measurement problem the proposed work addresses.
How should you measure ChatGPT visibility?
Define the test before looking for a win. Choose buyer-relevant prompts, decide which environments to test, and establish a repeatable collection process.
For each saved answer, record whether your brand was mentioned, recommended, or cited. Also record factual accuracy, the stated reason for inclusion, and the sources displayed. Keep the original response alongside the labels so another person can check your interpretation.
For a simple recommendation measure, divide the number of saved responses that recommend your brand by the total number of saved responses in that test set. Report the prompt set, dates, environments, and inclusion rule with the result. This is a measure of your collected sample, not every answer a buyer might receive.
Run a small repeatability check before building a larger reporting program. Repeat selected prompts under recorded conditions and compare the saved responses. Avoid presenting one favorable answer as a durable placement.
Keep platforms separate. If you test ChatGPT visibility, Gemini visibility, or Perplexity visibility, give each its own results. Record Claude tests and Google AI Overviews or AI Mode observations separately as well. Do not compress different interfaces and test methods into an unexplained “AI score.”
Connect the visibility report to business evidence you can collect. Track identifiable referral visits, relevant inquiries, and qualified opportunities. Ask prospects how they found you, but retain that answer as self-reported information rather than precise attribution.
Use attribution and analytics planning to define those categories before reporting results. Keep observed referrals, self-reported discovery, and sales outcomes distinct. Do not assign revenue to a brand mention without a defensible connection.
What mistakes should leaders avoid?
Do not assume absence means the business is a poor choice. First check whether the prompt describes your market, whether the answer contains errors, and whether your offer is clearly documented.
Do not publish unsupported superlatives, fabricated reviews, or invented customer outcomes to chase inclusion. Make the buying case with facts your team can defend.
Do not stuff the same phrase into every heading or create near-duplicate pages for every prompt variation. Give each page a clear job and link it to the relevant service or supporting explanation.
Do not buy a visibility report without asking how it was built. A useful report should show its prompts, collection dates, environments, saved responses, and classification rules. Ask how the provider handles incorrect facts and repeated tests.
Finally, do not accept a guaranteed recommendation as a substitute for a plan. Evaluate the work being offered: the diagnosis, the proposed changes, the evidence standards, and the reporting method.
What should you ask a GEO partner before hiring?
Ask for a sample diagnosis, not just a screenshot of a favorable answer. The provider should explain what the observation means, what it does not establish, and which action follows from it.
Request a scoped proposal. It should identify the pages involved, the buyer questions being addressed, the technical tasks, the sources available for claims, and the measurement plan.
Make factual correction part of the brief. Define who owns public company information and how inaccurate descriptions will be recorded and investigated.
The buying decision is straightforward: choose a partner whose proposed work is specific enough to inspect and whose reporting separates evidence from interpretation. That is more useful than chasing a supposed universal formula for brand recommendations.
Bring the answer you want to understand
Bring your buyer prompts, saved responses, and current service pages to a conversation about priorities. Start with the gap you want to diagnose, then discuss the work and measurement it calls for.
About Bob Generale
Bob Generale is President of Percepture.
