What do SEO, GEO and AEO mean?
SEO means search engine optimisation: helping search services find and understand your pages, and making those pages useful to the people who arrive.
GEO means generative engine optimisation. People use it for work intended to help a business or its content appear accurately in AI-generated answers. AEO, or answer engine optimisation, usually refers to making information answer questions clearly, across various answer services. Different suppliers draw the boundaries differently.
You do not need to settle the terminology before improving your site. Ask what a proposed job changes, where the improvement will be checked and what result would justify its cost. Google's own explanation keeps ordinary SEO foundations relevant to its generative search features. Google's approach
Which jobs should come first?
Make sure someone can establish what you sell, whom it is for, where you operate, the relevant price and conditions, and how to enquire or buy. Keep the facts current. A beautifully written page with an old service area can attract exactly the wrong customers.
Check that the important public pages work and can be discovered through normal links. Use Search Console or your website provider to investigate accidental exclusions, broken redirects or the wrong main page address. Test the page on a phone and try the actual next step.
Then answer the questions that real prospects ask before buying. Use a specific comparison, a useful example or evidence from your work where you can support it. Do not invent customer results, and do not turn one answer into dozens of pages by changing the wording.
For a local business, maintain the relevant profile’s category, opening hours, location and service details, and respond appropriately to genuine reviews. A shop should keep product facts and availability consistent with its feeds and sales systems. Choose the records your customers use; maintaining a profile on every platform can consume time without resolving an identified problem.
What changes when an assistant gives the answer?
An assistant can describe you using material from your site and elsewhere. Check the description as well as the appearance of your name. Does it understand the service? Has it repeated an old price? Is the cited source actually about the right business?
Save a small sample of genuine buying questions, with the service, date and market. Repeat those questions consistently when reviewing a change. One favourable answer is an observation, not proof that all prospective customers will see it. Use actual buying questions as well as the business name; flattering brand prompts alone do not test whether an unfamiliar prospect would find you.
Bing's guidelines describe SEO and GEO as connected work with different eligibility outcomes; they do not guarantee rankings or citations. That is a useful limit to put beside any supplier's promise. Bing's published guidance
How do you choose the right repair?
Consider this fictional mobile bicycle-repair business. Its website correctly says “Brighton call-outs from £45, parts extra”, but an old directory entry says “free collection across Sussex”. An assistant repeats the old directory claim and cites that entry.
Correct the directory record and check the public service details, then repeat the same bounded question sample. The correction needs checking: it cannot guarantee immediate replacement in every assistant answer. A new article about bicycle maintenance would leave the identified source untouched, and the sample still cannot show how often everyone else saw the error.
Suppose the booking page also has a broken mobile submit button. Repair and test the form. An accurate AI recommendation would still send someone into a failed journey. If the form works but the business cannot serve the requested postcode, make the service boundary clearer instead of treating the enquiry as a successful lead.
What can ordinary SEO checks miss?
Google has a separate Search generative AI control in Search Console, under Settings. Include is the default, so an owner normally has no new opt-in job. A property can instead be excluded or inherit a parent's choice. Check the effective choice for the property covering the relevant page. Google's control definitions
Extend the fictional bicycle business. Its service page is indexed, appears in ordinary Google results and has a working form. A supplier treats those checks as proof that the business is eligible for Google's AI features, then proposes another content rewrite because a small answer sample never cites it. Suppose the relevant Search Console property instead inherits Exclude from its parent.
That setting prevents the site's content and links appearing in the specified Google AI features without excluding it from ordinary Search. The source is the saved property choice, not an inference from the unfavourable answer sample. More articles or an llms.txt file would leave the exclusion in place. Ask the responsible owner whether it is intentional before changing it; if it is, retain the choice and adjust the visibility objective.
If the effective setting already says Include, there is no excluded control to repair or paid “AI unlock” to buy. Investigate the relevant pages, questions and representation instead. Inclusion is eligibility, not a citation guarantee. Google's help separates this feature choice from AI training and other Search ranking. The Google control cannot correct the directory claim repeated by a different assistant in the earlier example. Save the service, property, current choice and date so each proposed job addresses the system where the problem was found.
Do you need extra files or a new tool?
Only when they solve a defined problem at a sensible cost. Before buying an AI-visibility tool, decide what you will do differently with its report. Can your existing search reports, customer conversations and a small answer sample resolve the immediate question?
If a provider proposes llms.txt, Markdown or another technical addition, assess demand separately when costing that work. Ask which service uses it and what will be tested. Useful information, discoverable pages and a working enquiry route remain practical priorities.
Google Search does not require special AI files. For ChatGPT, OpenAI's OAI-SearchBot controls search access separately from GPTBot's training access. Ask for the particular access check you need, rather than a blanket change to all bots. Google's guidance, OpenAI's crawler definitions
How should you measure progress?
Keep a record of the pages or profiles changed and the problem each change addresses. Check the facts and customer journey after the repair. Then observe relevant search exposure, identifiable referral visits, suitable enquiries, retained orders and effort over a consistent period.
Count customers and payments separately from mentions and clicks. Missing source information stays unknown, and paid advertising needs its own category. Small numbers can be informative about individual problems without supporting a dependable growth percentage.
In GA4, Reports, Acquisition and Traffic acquisition is a practical place to start: inspect Session default channel group and the underlying source. AI Assistant covers recognised assistant referrals and excludes Google's AI Overviews and AI Mode. Keep a customer's remembered discovery with the existing enquiry or order so it is not counted as another customer. GA4's definitions
Use the weekly business review to choose the next job. The best early investment may be a corrected service boundary or a repaired form. Choose further SEO or GEO work when the evidence shows what it should improve and why that improvement matters to the business.