The DIY AEO Audit: What to Check Before You Spend a Dollar

Before budgeting anything for AI search, spend one structured afternoon finding out where you actually stand. This audit requires no tools beyond the engines themselves, a browser, and a spreadsheet, and it reliably surfaces which class of problem (visibility, accuracy, access, or content) deserves the first dollar.

Part one: the answer probes

Open ChatGPT, Perplexity, Gemini, and Claude. Run three groups of prompts and log the results verbatim.

• Brand probes: what is [brand], what does [brand] cost, is [brand] legitimate, who are [brand]'s competitors. You are checking accuracy and confidence: does the machine describe you crisply and correctly, hedge vaguely, or confuse you with someone else?
• Category probes: the five to ten questions a real buyer would ask when your product is the right answer, qualified the way buyers qualify them. You are checking presence: named, cited, or absent, and who wins instead.
• Source probes: for each category prompt, note every cited source. This list is your future target list, and it usually contains two or three sites you have never thought about.

Run each prompt more than once; answers vary, and frequencies are the honest reading.

Part two: the access checks

• Fetch your key pages without JavaScript and confirm your prices, descriptions, and Q&A exist in the raw HTML (the teardown covers why).
• Read your robots.txt and your CDN bot settings for accidental AI-crawler blocks.
• Confirm Bing indexes you; ChatGPT's retrieval has leaned on it, and many brands have never checked.
• Validate your structured data exists and matches the visible page.

Part three: the record checks

Assemble every public description of your brand (profiles, directories, aggregators, top brand-SERP results) and mark the disagreements: category labels, descriptions, locations, pricing. Then check the third-party layer for your category: are you present, correctly categorized, and decently reviewed on the platforms your source probes surfaced? Absence here explains absence in answers more often than anything on your own site.

Scoring and sequencing

Sort findings into four buckets. Access failures (blocked crawlers, invisible content): fix immediately, effects in days to weeks. Accuracy failures (wrong facts, entity confusion): fix second, via your own pages then the corroborating sources. Presence gaps on third parties: the campaign work, budget it over quarters. Content gaps (prompts where every cited source is generic): your editorial roadmap, one strong page per question.

Then make it repeatable: fix the prompt set, schedule the probes, track the trend, per the measurement method. The audit is diagnostic, not treatment, and the timeline piece sets expectations for each fix class honestly: an afternoon of looking, so the next quarter of work aims at the problem you actually have.