ChatGPT visibility audit template

Start with columns for prompt, prompt family, audience, scenario, answer status, brand position, named competitors, visible source links, uncited source mentions, factual claims, wrong or unsupported facts, and recommended next action. This structure turns a ChatGPT visibility audit spreadsheet into evidence instead of a collage.

Add a decision-quality field. Mark whether the answer gives a buyer a reason to shortlist the brand, compare it, contact it, avoid it, or keep researching. That field prevents vanity reporting because a passing mention and a purchase-relevant recommendation are not the same outcome.

Use prompt families so the audit is broad enough to reveal patterns. For example: “best tools for enterprise customer onboarding,” “alternatives to [competitor],” “which agency should a B2B SaaS company use for AI visibility,” and “compare [brand] with [competitor] for mid-market teams.” The forced brand prompt is useful only for factual accuracy review.

Keep the wording neutral. Avoid stuffing the brand name into every prompt or asking leading questions that manufacture a favorable answer. A useful ChatGPT recommendation audit template separates discovery prompts, comparison prompts, branded prompts, and correction prompts so the final report can explain which kind of visibility was tested.

A clean capture row should include: run ID, date, model or product surface as shown to the user, search on or off, prompt, full response, sources shown, source URLs, screenshot file path, recommendation classification, factual issues, and reviewer initials. This is the minimum audit trail for a defensible report.

Source capture matters because ChatGPT answers may include visible links, source panels, or no links depending on the experience. The audit should distinguish “answer said this,” “source visibly supported this,” and “reviewer verified this elsewhere.” OpenAI’s own accuracy guidance makes citation checking important because sourced-looking answers still need verification.

Use strict definitions. “Recommended” means ChatGPT presents the brand as a suitable choice for the prompt. “Shortlisted” means it appears among viable options. “Compared” means it is evaluated against alternatives. “Cited only” means a brand source supports an answer but the brand is not the recommended solution.

Add severity for errors. A wrong founding date may be low severity, while a false pricing claim, discontinued product claim, or misfit category label can materially damage buyer perception. The template should assign each wrong fact an owner, public-source fix, and status so the audit leads to repairs.

The executive summary should separate visibility from accuracy. A brand can be visible but misdescribed, absent but well documented, or cited without being recommended. Those are different problems. The report should show examples from the spreadsheet and summarize patterns by prompt family, not by the most flattering screenshot.

The repair section should map each issue to public evidence: missing comparison pages, weak category pages, outdated profiles, unclear pricing pages, thin case studies, inconsistent third-party descriptions, or unsupported claims. Google’s crawler documentation reinforces the practical point: answer systems need discoverable web evidence before they can reliably reflect a brand.

Do not claim that the audit reveals how ChatGPT ranks brands internally. It does not. It shows what happened under documented conditions. A disciplined template borrows from measurement practice: define scope, record conditions, repeat important prompts, preserve evidence, and separate observation from interpretation.

Outreach needs extra restraint. FTC advertising guidance makes clear that marketing claims should not mislead. Do not tell a prospect they are “invisible everywhere” because one prompt missed them, and do not imply guaranteed placement after fixes. Use the screenshot as a conversation starter only when the underlying audit supports the claim.