SEO and GEO: what should you fix first?

Begin with a question your customers actually ask

SEO concerns how people discover and use your pages through search. GEO, short for generative engine optimisation, commonly describes work on visibility in AI-generated answers. Those labels can help organise a project, but they do not tell you which page deserves the next edit.

Start with a real decision: does this service cover my location, does this product fit my needs, or what happens after I connect my repository? A useful page answers the question directly and supplies the evidence behind its answer. Repeating a keyword across several shallow posts gives the reader less help than one complete explanation.

For Your Optimiser, that means explaining what runs autonomously and where review remains in the workflow. Publishing our real repair method is more informative than claiming an unexplained “AI visibility score.” The same principle applies to your own product: explain what it does, where it works and what it cannot establish.

Resolve access and page integrity before adding more copy

A helpful explanation cannot do much for a visitor who follows a broken link. Check the public URL, important internal links, preferred URL declarations and whether the intended content reaches the served page. Preserve private areas. Crawl access and indexing are distinct: permitting a fetch is not proof that a page is indexed.

An AI-answer result also depends on the engine and retrieval context. A page that is available to one crawler is not automatically used by every answer system. Avoid a blanket “allow all bots” change when a site owner has deliberately chosen a crawler policy.

Source: Google’s explanation of robots.txt and its limits ↗

For Google, special AI files are not the starting point

Google’s current guidance for its generative Search features points back to established SEO and useful, distinctive content. It does not recommend llms.txt or special AI markup as a way to improve visibility in Google. That is a Google-specific statement, not a universal claim about every consumer of an llms.txt file.

Our own llms.txt is a readable index of product facts and public links. We do not treat its presence as a ranking improvement or use it to control crawlers. Likewise, article markup describes an article; it cannot turn unsupported copy into trustworthy evidence or guarantee an AI citation.

Source: Google’s generative AI search optimisation guide ↗

Make your claims specific enough to check

Add original material where it helps the decision: a documented repair, a real product specification, a clear explanation of your process, or a service limitation. Keep business identity, pricing and availability consistent across relevant pages. If those facts change, update the explanation and its supporting records together.

Structure the page for readers. Use a descriptive heading, a direct introduction and sections that answer the next likely questions. Explain the method and limits of any result. A fictional demonstration should say it is fictional; a local test should not be presented as a customer’s production improvement.

Do not pad a page to reach an arbitrary length. A blog is a publishing format, not a ranking feature. Before creating another article, ask whether it solves a different reader problem or only restates what an existing page already covers.

Source: Google’s people-first content guidance ↗

Keep repair, visibility and value in separate columns

A verified repair answers whether a known technical defect was resolved. Search impressions and clicks describe search exposure and visits. An AI citation tells you that an answer referenced a source in an observed response. An enquiry or purchase is a business event. None of these is a replacement for the others.

For an AI-answer experiment, save the engine, question, date, whether retrieval occurred, cited URL and whether the answer represented the page accurately. Repeat the observation and keep cases with no citation. Record any changes made between observations so a surprising answer is not mistaken for a stable trend.

Prioritise a confirmed broken journey ahead of a speculative optimisation trick. Once the fundamentals work, use your visibility observations to decide which explanations need more substance. That gives both readers and your team a reason for the next edit.