Work through AI Visibility and GEO Playbook for Small Businesses

Suppose a homeowner asks an AI assistant which local company can answer a garage-door repair call after hours. Your website might explain that service well. The next question is observable: did the assistant mention your business, link to your page or recommend you?

This playbook helps you improve the page and keep a record of what actually happens. It works for a trade, law firm, clinic, accounting practice or another small business. Start with an existing page and a real buying question.

Keep three records separate

TQP's AI Visibility Score is an automated website-readiness checklist. It reads sampled HTML for text and markup signals. A score of 80 does not mean an 80% chance of recommendation. Review its evidence before changing good content.

Pick questions that sound like customers

Choose a small, repeatable set from sales conversations, search data and the services you actually sell. Include the job, the location where relevant and the purchase constraint. Keep brand names out of the discovery questions.

These are question examples, not measured searches. Replace the location and service with your actual scope. Ask broadly enough to reflect a real decision, then include specific questions where your business has a useful answer. Keep branded checks in a separate group.

Inspect the page a buyer would land on

Read the whole page before adding another FAQ. Can someone tell what you do, who it is for, where you work, what the first step involves and how to reach you? Show the important distinctions: an estimate request versus a confirmed visit, a consultation versus accepting a legal matter, or an assessment versus a treatment booking.

Use prices and inclusions you can stand behind. Show your people, work and useful demonstrations where appropriate. Keep sources beside claims that depend on them. Explain why you are the right choice with specific service details and evidence a buyer can inspect.

Write down one concrete improvement. For example: “The repair page never says whether the office handles evening calls. Confirm the real coverage, explain the next step and test the phone link.” Save the original page and the change date. Keep what already helps the customer.

Check access before adding technical extras

Ask your developer to confirm the public page loads, the important content is readable, internal links lead to it, and its canonical and crawl settings match the intended page. Check the actual returned HTML and relevant webmaster reports, not just a checker badge. A failed request from one client is a clue to investigate, not proof that every search bot is blocked.

Google says AI Overviews and AI Mode use the same foundational SEO requirements. A supporting page must be indexed and eligible for a search snippet. Read the property's Search generative AI setting in Search Console as well: an excluded site is not eligible for the covered AI features. See Google's inclusion control. Google does not require special AI text files or special schema for these features. Structured data should agree with visible content. See Google's AI features guidance.

For ChatGPT Search, OpenAI distinguishes OAI-SearchBot from GPTBot, which serves a training purpose. Review the crawler and network settings for the intended use rather than treating those controls as interchangeable. See OpenAI's crawler documentation.

Keep useful existing discovery files accurate. Do not treat their presence, a HowTo block or a perfect checker score as evidence of recommendations.

Record the actual answer

Use the search platform itself when accessible. Keep the wording and test conditions consistent. Start a fresh conversation and avoid telling the assistant your brand or asking it to recommend you. Record the account context, including whether it belongs to your business; a fresh chat does not by itself prove the result is independent of personalization.

Count a brand mention, a linked citation and a recommendation separately. A source link may appear without an endorsement. A recommendation may name a business without linking its site. If the platform is unavailable, record that limit; do not fill the row with a model-written imitation.

Learn from the cited pages

Open the sources that actually support the answer. Note the passage the platform could use: service availability, a price and its limits, a useful comparison, an original example, a working tool or a credible outside reference.

Then compare that passage with your existing page. Is yours missing the answer, burying it, contradicting another offer or making the next step difficult? Fix the demonstrated gap. A competitor's claim is evidence of what its page says, not proof that it delivers better results or that copying its format will earn a citation.

Measure enquiries alongside visibility

Use one definition of a qualified enquiry: a real prospect asking about a service you provide, in an area you serve, with a need you can reasonably handle. Exclude spam, job applications, vendor pitches and duplicates. Record useful conversations even if they do not close immediately.

Keep each platform separate. Report the numerator and denominator, not just a percentage. An incomplete test is not a zero. Referrers can be missing, so label self-reported discovery separately from a recorded referral. Do not expose client or patient details in a public tracker.

A practical review rhythm

1. Save the baseline, choose the important questions and identify the existing pages that should answer them.
2. Fix access or buying-path failures first. Then improve the most useful missing passage, demonstration or comparison. Record the release date and what changed.
3. Review qualified enquiries as they arrive. Recheck the same AI questions on a planned monthly cadence rather than after every wording edit.
4. Compare complete, equal-length search and enquiry windows. Note seasonality, promotions, other site changes and small samples before interpreting movement.
5. Decide what to keep, repair or deepen. Expand to another page only when it has a distinct customer job and enough useful material.

Google now provides a Generative AI performance report in Search Console. Export it separately from the normal Web report. Keep the property, search type, dates and page or country filters with the export. It measures impressions in the covered Google AI features, not an endorsement or a qualified enquiry.

Do not add those impressions to Web totals or sum page rows to reconstruct the property total: the reports overlap and their aggregation differs. If the AI report is unavailable, record that limitation rather than entering zero. Low-impression properties may not see it.

Use the native report to identify pages receiving visibility, then use recorded answers to inspect the wording and sources. Do not label all organic clicks as AI referrals. Search and response changes remain observations unless you have evidence that isolates their cause.

Put the finding to work

Choose TQP when you want the website, enquiry route or connected customer system designed and implemented for you. Bring the page, the recorded answer and the enquiry problem to a free consultation. We can scope the work around the gap that matters, starting with an affordable website, local search, AI answering or booking and intake.

Sources checked October 22, 2026. The examples and review rhythm are TQP's practical guidance, not customer results or promises of ranking, citation or lead volume.