AI Search SEO: A Practical Guide to AI Overviews and AI Mode
A practical AI search SEO audit for business websites: indexing, Google’s AI controls, useful content, author identity, image discovery, and measurable results.
Full-stack web developer · Cairo
AI search SEO starts with a page worth finding, not a file called llms.txt. For a business website, the useful question is: can search engines access the page, understand the answer it provides, and connect it to a credible person or company?
This guide was checked against Google’s documentation on September 10, 2026. It includes an important distinction between technical eligibility and actual visibility. The audit examples below are proposed working methods, not a claim that this portfolio or a client has earned a particular AI citation.
What changed—and what did not
Google’s current generative AI optimization guide places AI search within SEO, rather than treating it as an independent ranking shortcut. It says there is no special schema or AI text file required for Google Search. It also advises against mass-producing pages for every possible variation of a question.
That changes how I would evaluate an SEO proposal. “We will publish fifty AI-optimized answers” is an output, not a business result. “We will make your product comparison usable, document its evidence, and measure whether it attracts qualified enquiries” is a much clearer project.
AEO and GEO are common labels for answer-engine and generative-engine optimization. They can describe useful work, but the label does not tell you whether the work is valuable. Ask to see the actual page improvements and the measurement plan.
Start with eligibility: one URL, four checks
Take the exact URL you want people to discover. Do not begin by testing the homepage if the problem is an individual article.
| Check | What to inspect | What a failure means |
|---|---|---|
| Access | Final response, redirects, login barriers, crawler restrictions | The intended page may not be available to a crawler |
| Indexing | URL Inspection and Google-selected canonical | Google may not have indexed this URL, or may prefer another |
| Content | Rendered title, main answer, links and images | Important information may be missing or difficult to interpret |
| AI inclusion | Search Console’s Search generative AI setting | The property may be excluded from those experiences |
Google documents a Search generative AI control under Search Console Settings. Inclusion is the default, but a property can inherit a parent property’s choice. Check the effective setting instead of assuming a recently created URL-prefix property overrides the domain. Inclusion allows participation; it does not promise selection.
For a Next.js site, I would also request the page without a browser and inspect the returned HTML. This is a useful engineering check that the title, main article and internal links exist before interactive components start. It is not a claim that Google cannot process JavaScript. The implementation details belong in my Next.js technical SEO guide.
Turn a broad topic into a decision someone can make
Imagine a restaurant owner searching for an AI assistant that answers menu questions. A page titled “The Future of AI for Restaurants” can be polished and still leave that person with no decision.
A more useful article would answer: where do menu prices come from, how quickly do changes appear, what happens if an allergen answer is uncertain, and when does a staff member take over?
I would build its editorial brief around five pieces of evidence:
- 1.A specific audience: an owner or manager maintaining a frequently changing menu.
- 2.A concrete decision: whether a website assistant is suitable for the questions customers ask.
- 3.A worked example: a price changes at noon; trace how the assistant obtains the updated value.
- 4.A failure boundary: the assistant must not guess about allergens or present generated dietary advice as verified fact.
- 5.A maintenance responsibility: identify who owns corrections, escalation and the authoritative menu.
These are proposed editorial requirements, not a restaurant case study. If there is a real deployment, replace assumptions with permissioned screenshots, actual constraints and dated measurements. If there is not, label the design as hypothetical. A fabricated success story would weaken the trust the article is supposed to create.
The same method works for a portfolio. An article about a deployment failure can be useful if it shows the configuration, the observed symptom, the diagnosis and a reproducible check. A generic list of technologies cannot substitute for that explanation.
Make authorship clear without repeating your name everywhere
A reader should be able to answer who wrote the article and why that person is relevant to the subject. Use a visible byline linked to a substantive profile page, a short accurate biography, and links to the author’s real professional accounts.
Google’s ProfilePage documentation describes markup for pages focused on a person or organization. Use it on an appropriate profile page, and keep the marked-up identity consistent with what visitors see. It is not a mechanism for claiming somebody else’s profile or purchasing a knowledge panel.
For this kind of portfolio, my identity checklist is straightforward: one spelling of the name, one canonical profile URL, the correct LinkedIn account, and a biography that does not inflate experience. An author image should depict the author. If a portrait is AI-edited or used in a conceptual illustration, describe it honestly rather than presenting an invented event as documentary photography.
Help people discover images, not just filenames
A descriptive filename is useful housekeeping, but it cannot force a portrait into a name search. Put meaningful images into actual page content, with relevant nearby text and accurate alternative text. Do not turn alt text into a list of keywords.
Google’s image SEO guidance covers discoverable HTML images, image quality, contextual information and image sitemaps. Apply those fundamentals to the author portrait and article artwork. Keep image URLs reachable, and avoid unnecessarily removing an older public image URL during a redesign.
My release check is practical: open the image directly, verify its response type, check the mobile crop, and make sure the author’s face has not been cut off. Then check that the article’s structured data describes the image actually shown. Decorative artwork and a headshot have different purposes; using the same face on every paragraph does not add useful information.
Measure visibility separately from business value
The Generative AI performance report documents impressions for AI features, with dimensions including pages, countries, dates and devices. Google notes that insufficient impressions can explain why the report is unavailable. A missing report alone is not evidence that your site has a technical error.
I would maintain two small scorecards. The discovery scorecard tracks indexed URLs, search impressions and the pages receiving them. The business scorecard tracks meaningful actions such as a relevant enquiry, a booked call or a completed purchase. A page can improve on one scorecard without improving on the other.
Record the publication date and the nature of each substantive edit. Compare equivalent periods where possible, and watch for seasonality, brand campaigns and unrelated site changes. With very little traffic, one additional enquiry can swing a percentage dramatically; keep the raw counts beside percentages.
Do not describe an increase in overall organic traffic as proven AI traffic unless the measurement actually supports that attribution. Equally, an AI citation screenshot is not evidence of a stable ranking or a conversion.
A manageable publishing and review plan
Before publishing: choose one primary reader question, verify factual claims, add supporting links, and remove sections that do not help the decision. Check the page on a phone.
At publication: verify the canonical URL, indexing directives, author links and image responses. Add a relevant internal link from an existing article and list the new URL in the sitemap. Google’s sitemap guidance makes clear that a sitemap helps discovery but does not guarantee crawling or indexing. Use modification dates for meaningful changes, not an automatic daily freshness stamp.
After publication: inspect indexing status and collect the questions readers still ask. Revisit the article when its underlying advice changes. Expand a missing explanation on the existing page before creating a nearly identical article.
Frequently asked questions
Do I need llms.txt for Google AI Overviews?
No. Google’s current optimization guidance does not require it or treat it as a Google Search ranking advantage. Maintain one only if another service you intentionally support has a useful reason for it.
Can structured data guarantee my photo or social profiles appear?
No. Structured data can clarify supported information, but search engines choose the results and presentation. Keep the real portrait, visible author information and official profile links consistent.
What should I fix first on a small business website?
Fix a genuinely blocked or broken important page first. Then improve the page that best answers a commercially relevant customer question. Publishing more articles should not be a substitute for either task.
If you need the implementation side, explore my selected web projects or get in touch about your website. The goal is a useful, discoverable site that earns trust—not a promise to occupy every search result.