SEO / GEO PLAYBOOK

Make your business easier for AI to understand

Generative engine optimization starts with making your business understandable, not with a special file or a promise of AI visibility. A useful foundation combines accessible pages, consistent facts and content that answers real questions. Search engines and AI products differ in how they discover and cite information. None of the steps below guarantees inclusion, a citation or a ranking.

Make the essential pages accessible

Begin with ordinary technical hygiene. Important information should be available in the initial HTML, linked from sensible navigation and reachable without signing in. Use descriptive page titles, real headings and useful link labels. Check the HTTP status of every important URL. A visually polished page that returns an error or hides its main content behind an interaction is a weak starting point for both visitors and crawlers.

Publish a sitemap containing canonical, indexable URLs. Use robots.txt intentionally, understanding that crawler instructions are not access controls. A blocked path can still be known from other links. Protect private information with authentication instead of relying on exclusion rules. Review the current documentation for each search or AI crawler before changing its access; their policies and user agents are not interchangeable.

Describe the same business consistently

Your name, legal identity, services, contact details and geographic context should agree across your pages. Separate a founder’s professional experience from work delivered by the company. Name an in-house project as such. Avoid vague claims that force a reader to guess what you actually provide. A clear service description should explain the problem, the deliverable, the boundaries and the next step.

Use structured data to describe those visible facts. Organization, Service and BreadcrumbList can make relationships explicit. Article markup can identify a genuine author and publisher. Markup should match the page rather than introducing unsupported reviews, ratings or qualifications. Validate its syntax and the applicable requirements, but remember that valid JSON-LD is a description, not an entitlement to a search feature or an AI citation.

Answer a question completely

Choose questions you can answer from actual knowledge: how a webhook is verified, what a website migration includes or when an automation should stop. Lead with the answer, then explain the trade-offs and the steps. Define unfamiliar terms where they first appear. Examples should be labelled as examples, especially when they resemble a client result. Specific instructions are more useful than a paragraph repeating service keywords.

Link to primary documentation where it helps someone verify a technical statement. Distinguish a current implementation detail from an opinion or a recommendation. Do not add a date merely to make old content look fresh. When software behaviour changes, review the affected instructions and update the relevant publication information honestly. Content maintenance is part of the system, not a one-time search exercise.

Treat llms.txt as an optional reading aid

The llms.txt proposal offers a Markdown-style introduction to a site with useful links. It can help a person or tool find an organized summary, but it is not a universal discovery standard. Support varies, and publishing it does not require any AI product to read, index or cite your site. Keep it small enough to understand and point to the canonical source pages.

A companion llms-full.txt can provide longer text for convenient reading. Generate it from the same content as the visible site so the two do not drift apart. Do not put private information, hidden instructions or claims you would not publish on the page into either file. These files are public resources. They do not replace structured data, a sitemap, access control or useful human-readable pages.

Keep bilingual and duplicate content coherent

Give English and French pages their own stable URLs. Use a canonical URL appropriate to each language and reciprocal hreflang alternates. Translate meaning rather than swapping isolated labels while leaving the body in another language. A visitor following a language switch should land on the equivalent page, not lose their place at the homepage. Consistent internal links help keep those relationships understandable.

Avoid creating many near-identical pages merely to target different phrases or locations. If a page does not offer a distinct answer, consolidate it. For genuine variations, make the differences useful and visible. Check that a redesign preserves important URLs or supplies appropriate redirects. Neither an AI summary file nor extra schema can repair a collection of confusing, duplicated or missing source pages.

Evaluate accuracy before chasing mentions

Create a small review list: can a visitor identify the business, understand each service and verify the important claims? Can a crawler retrieve the intended page and its metadata? Are source links current? Parse structured data in automated tests and inspect rendered output after content changes. These checks are repeatable and actionable, unlike a promise that one prompt will always produce the same AI response.

If you observe AI answers, record the product, date, prompt and cited URLs. Results vary by context, so use them as observations rather than proof of coverage. Correct errors at the source and improve the underlying explanation. The durable aim is a site that is easy to read, maintain and verify, whether the next reader is a prospective client, a search crawler or an AI system.

AGWritten by Andy Guitteaud · GingerClick
Practical guidance, not legal advice. Adapt it to your environment.

A system that fits your business starts by understanding it.

A first conversation to understand your business and identify what is worth building.