AI answers

SEO and GEO: a practical starting point for a small business

Updated · Publisher: AMA CLASS, K.K.

Short answer

Start with a page customers read before contacting you. Check that search engines can access it, answer the questions that help someone choose, and record each change. GEO focuses on visibility in AI-generated answers. Much of that work overlaps with SEO, and neither guarantees a recommendation.

Check the supported features and availability: Features

1. Choose one useful page, not a publishing quota

Pick a service or product page that a prospective customer would visit before contacting you. Write down the customer question it should answer, such as whether your service covers their area or what is included in a quote. Save the current wording and URL before changing anything.

Before adding an article, check whether an existing page leaves an important customer question unanswered.

2. Check what the live site actually sends

Ask the person who maintains the site to compare the HTML returned by the server with the page people see. Google can render JavaScript, so an HTML response with little content is not automatic proof of invisibility. Some other crawlers cannot run it. Serving important text directly can make that content easier to access.

Check the live robots.txt as well as hosting and firewall rules. A successful request made with a bot name is only a diagnostic clue: it does not prove that the real bot is allowed. Inspect server logs and the provider’s verification guidance. A planned 301 redirect to the main address is not an error; follow it and check the destination.

3. Keep search access separate from model training

OpenAI uses OAI-SearchBot for search and GPTBot to collect content that may be used for model training. The two settings are independent. Decide which use you permit instead of allowing every bot because a checklist says so.

Allowing access is not a promise of inclusion, citation or recommendation. Keep private pages private. If a security rule appears to block legitimate discovery, ask for a narrowly scoped correction rather than switching protection off for the whole site.

4. Make page signals agree

Review the intended public URL, internal links and canonical URL together. The canonical tells search engines which version you prefer; it should not accidentally name a different service. Check for an unintended noindex instruction before adding more content.

A sitemap can help discovery, although Google does not require every site to have one. Keep it focused on the public pages you want indexed. Structured data should describe what is actually visible on the page. Do not invent ratings or add a software schema to a business that does not sell software.

5. Answer the question behind the purchase

A useful comparison starts with a real choice: one-off maintenance or an ongoing plan, an in-house task or outside help. Explain who each option suits, what is included, what is excluded and what the customer must prepare. Include limits of your own offer.

For a maintenance service, explain which updates are included and which require a separate quote. A screenshot of the request process or a documented test can make the explanation clearer. Use evidence you actually have; do not invent a customer story to make a page sound convincing.

6. Keep facts consistent beyond your own site

Maintain a dated reference sheet with your business name, service area, offer and contact details. Use it to find contradictions in your own pages and legitimate business listings. Change a fact because it is wrong, not because an AI answer used different wording.

If an AI response makes a wrong claim, record the question, date and cited page before investigating. Answers vary, and the cited page may not explain every part of a response. Correct the source you control and check again later. Avoid purchased ranking links, fabricated reviews and automated link-exchange schemes.

7. Measure discovery and business outcomes separately

Use Search Console to see Google search activity and Bing Webmaster Tools for Bing’s own data. In your records, separate a brand mention, a source link, a visit and an enquiry. None proves the next happened.

Repeat a small set of genuine buyer questions under recorded conditions if you test AI answers. Keep the wording, date, language and cited URLs. Small samples are observations, not market-share estimates. A voluntary “How did you hear about us?” question can add context, but memory and attribution are imperfect.

8. Make today’s change safe and reviewable

Let an AI assistant read evidence and draft a change first. Ask it to show the exact page, proposed edit, supporting source and verification step. Review changes to robots.txt, canonicals, redirects and noindex before they are applied. Keep a backup and a way to restore the previous version.

Today, choose one unanswered question on the page and address it with facts you can verify. Then check the published text and contact process on a phone. Record the date, but do not attribute the next day’s search performance to that edit alone. Explore get1pct if you would like a more manageable way to approach website improvements.

How to use Search Console: choose a page without misreading CTR

AI search and your website: what Google says, and what to check first

Related fundamentals and practical guides: SEO, AIO and GEO glossary: why SEO matters and what to check / Free PR sites by country: press releases and product listings

Search and visits for anonymous site A

Measured data for one anonymous site, comparing two non-overlapping 28-day periods.

Metric2026-09-14– 2026-10-112026-10-12– 2026-11-08
Google Search impressions3,440,4614,243,412
Google Search clicks120,039147,282
Google Search CTR3.49%3.47%
Site-wide average position5.454.99
GA4 Organic Search (sessions)158,762191,813
GA4 AI Assistant (sessions)175243
Our API-derived aggregate, retrieved November 11, 2026, anonymized and reconstructed as a table; it is not a screenshot of the original interface.
Search Console: America/Los_Angeles (PT) · GA4: Asia/Tokyo (JST)

Clicks rose about 22.7%, while CTR was nearly flat. Do not judge success by CTR alone; narrow the report by query, country and device.

This observational comparison does not establish that any specific change or service caused the difference or guarantee results for other sites. GSC Google Search clicks and GA4 sessions are different measures; their search engines, measurement methods and time zones differ, so totals do not match. AI Assistant excludes Google AI Overviews and AI Mode and does not count all AI use or citations.

Sources