Start with the decision a buyer needs to make
A homeowner looking for a weekend repair and a procurement lead comparing retainers need different facts. Pick one audience, one offer page and one action, such as requesting an estimate. Write down which facts would change that person’s choice. Price basis, coverage area, turnaround and exclusions are good candidates.
Answer engine optimization (AEO) means helping answer systems find and explain those facts. Generative engine optimization (GEO) also considers how a generated comparison presents your brand and its sources. A buyer can receive a useful answer without clicking. A store or service business still needs to know whether the right buyer reached the next step.
Days 1–3 · Record a baseline you can repeat
Save the exact wording of three questions. For discovery, try “Who can help a small service business improve how it appears in AI search?” For comparison, try “Which providers publish the price and scope of an AI search audit?” For a fact check, name your business and ask “What does the review include, what costs extra, and where do I start?” Keep branded and unbranded results in separate rows.
Use two assistants your buyers use. Start a fresh conversation for each question, enable the same web-search setting, and run each question twice. That gives 12 planned observations. Record the date, account or temporary-session state, location, model or mode when shown, the exact answer and its linked sources. Mark blocked or unanswered tests separately. You can use the same sheet for Muse, Dot or Grok if available to you. Record their actual mode and any access to saved brand context.
Download the editable CSV audit log. Copy a row for each assistant and repeat. Save answer evidence privately and use aggregate customer counts.
Days 4–10 · Find the fact that would change the answer
Read each linked source. Does it support the price, scope and next action the assistant stated? Mark each as correct, incorrect or unknown. Identify the first consequential error. If a tool quotes an old retainer price for a one-time service, trace it to the specific page, directory or cached comparison. Fix the page you control and request a factual correction where another publisher owns the listing.
Open the offer page while signed out. Check its status, canonical URL, indexability, links from your site and whether the facts are present as readable text. Google requires an indexed page eligible for snippets for inclusion as a supporting link in AI Mode or AI Overviews. OpenAI documents separate choices for OAI-SearchBot and GPTBot. Review search access independently from your training preference.
Days 11–20 · Publish one fix and test the handoff
Replace an ambiguous sentence with a buyer-facing answer that identifies who qualifies, what is included and how to start. Link to the underlying evidence and the actual action. Keep a dated copy of the old wording and the release date. Give a colleague the original question and the updated URL. Can they quote the right facts and reach the booking or checkout screen? Record exactly where they stopped.
Days 21–30 · Score the answers and the business outcome
Repeat the same questions, modes and number of runs. For each assistant, divide answers that explicitly recommend you for the stated need by completed discovery answers. Calculate owned-page citations separately. A mention counts when your name appears. A recommendation requires the answer to say you fit the request. A citation requires a link. A wrong price still counts as a factual error even if you were recommended.
For example, 2 recommendations in 8 completed discovery answers is 25% of that sample. If the page also received 3 tagged visits and 1 qualified inquiry, report those as separate counts. Compare equal reporting windows, log other campaigns and use “unknown” where referrer data is missing. With small samples, keep the counts alongside the percentages. Repeated wrong facts are a reason to revise the page even when traffic is too small to interpret.
Put it to work
- Save the buyer questions and the original offer page.
- Publish one factual correction and record its URL and date.
- Repeat the sampled questions and review qualified inquiries in a matching time window.
Sources and further reading
Sources reviewed November 19, 2026. Platform behavior and availability can change. The exercises and examples are our editorial recommendations.