Define the buyer scenarios
When I look at AI visibility, I start with the customer. What are they trying to decide, and can they find a clear, useful answer about your business? Build a small prompt set across five categories: category discovery, problem diagnosis, product or provider comparison, risk reduction, and action. A retailer might test “best sectional for a home with pets,” while a consulting firm might test “who can audit a retailer’s visibility in ChatGPT?”
Record the audience, location, decision stage and intended action for every prompt. That keeps the work tied to customer and commercial value.
Test technical access
Before judging the answers, confirm that important pages can be retrieved. Review robots.txt, status codes, canonicals, sitemaps, server-rendered text, CDN and WAF controls, and bot rules. Test from the live production edge, because a correct source file does not prove the hosting layer serves the same response.
OpenAI separates OAI-SearchBot, used for ChatGPT search inclusion, from GPTBot, used for potential model training. Those choices should be intentional. Also confirm that important content remains visible without relying on a browser to execute client-side JavaScript.
Run a controlled baseline
Use fresh sessions and fixed conditions. Repeat prompts to see how much the answers vary. Three runs can be a useful starting point, but they are not a statistically reliable estimate of overall visibility. Capture whether the brand appears, the position and context of the mention, accuracy, cited URLs, named competitors, recommended next action and any material omissions.
Do not reduce the result to a single vanity score. A brand can appear frequently but be described inaccurately, or receive citations without being recommended. Track inclusion, accuracy, citation and competitive share separately.
Trace the evidence
For each answer, ask what information supports the recommendation. A citation tells you where an answer points; it does not prove that every claim in the answer is correct. Review dedicated service or product pages, case stories, author pages, policies, reviews, structured data and independent coverage. Look for contradictions in names, descriptions, dates, prices, capabilities or locations.
I want the findings to show what is clear, what is missing, and where a competitor makes the buying decision easier. Those are more useful priorities than a score with no explanation.
Turn findings into a roadmap
Prioritize a small number of high-intent pages and evidence gaps. Assign owners for technical access, content, structured data, product data, third-party authority and measurement. Confirm indexing and measurement before expanding page volume.
Repeat the baseline on a consistent schedule. You can't control what an AI says on any given day. You can make your business easier to describe accurately, and measure whether that's improving.
Verification
Linked sources
These links allow readers to review selected original and supporting material.
About the author
Jeff Douglas
I bring 20+ years of ecommerce leadership experience, including revenue and budgeting responsibilities, customer experience and digital marketing leadership, Salesforce Commerce Cloud modernization and high-consideration retail. I advise businesses on customer experience, growth priorities, technology and AI-driven discovery.
Experience