A website being accessible and readable does not explain why an AI answer selects it as a source. Receiving a page and using it to answer a question involve different decisions. Before commissioning another redesign or twenty new articles, identify what you need to improve: access, available information, relevance to the question or the observed source selection.
This guide helps you prepare that decision. It includes an original diagnostic map, a real public sample and a worksheet for commissioning the next intervention. If you lack a consistent comparison of questions, pages and cited sources, start with an LLM visibility audit. A bounded assessment helps put the budget behind an observed gap.
Define what “not appearing” means
Four situations commonly get mixed together: the brand is absent, it is mentioned without a link, an owned page is cited without recommending the business, or the description is inaccurate. Record which you observed. Being cited for a technical explanation is different from entering a supplier shortlist.
Identify the product, search mode, exact question, language, market, date and conversation context. Retain the answer and displayed links. A branded question and a general provider recommendation test different tasks. Diagnosis needs a reference you can repeat.
Choose one commercially important question and the page that should help answer it. A buyer asking which supplier integrates a technical catalogue with a distributor portal needs evidence of that capability. Publishing company news about artificial intelligence does not establish that the page supports this purchase decision.
Locate the layer before buying a solution
| Layer | Question to resolve | Useful evidence |
|---|---|---|
| Access | What does the relevant request receive? | Rule, response and request context |
| Availability | Is important content present in the received source? | Passage, version and exact URL |
| Relevance | Does it answer the question’s buying decision? | Intent compared with page content |
| Corroboration | Which facts and sources support the claim? | Origin, authorship, date and evidence boundaries |
| Selection | Which sources appear in the observed answer? | Archived answer, links and context |
These layers organise an investigation. They are not a claimed formula for every engine. A successful check at one layer does not settle the next. Reading a page supports examining its content; establishing a citation requires observing the corresponding answer.
Check access for the particular product
Your browser and a crawler can receive different responses. Review robots.txt, infrastructure rules and available logs with the technical owner. Writing a User-Agent header in your own tool does not authenticate a request as a genuine crawler.
Official OpenAI crawler documentation separates OAI-SearchBot, used for search, from GPTBot, associated with content that may be used for training. Those controls are independent. Training access should not be imposed as a prerequisite for commissioning search-visibility work.
Ask for the relevant rule and its purpose to be reviewed. If protection blocks a legitimate request, define a narrow adjustment and verification. Preserve security controls and deliberate access decisions. Opening the infrastructure indiscriminately does not answer the diagnostic question.
A public example: declared access and received content
On 25 October 2026 at 03:32 UTC, we made one ordinary HTTP request to robots.txt and another to The GEO Company’s technical service page. Both returned 200. We retained the files and checksums. We did not authenticate as a crawler or inspect logs, Google’s index or AI answers.
| Reference | Actual observation | Next distinction |
|---|---|---|
| A01 · robots.txt | Explicit OAI-SearchBot group with Allow: / | An access declaration does not establish a crawler visit |
| A02 · Technical service | 200 and service definition present in initial HTML | Receiving text does not establish selection or citation in an answer |
The sample rules out an empty page for that ordinary request and a missing specified group in the observed robots file. It does not establish identical treatment for every visitor. We have not presented an actual absence of citations for The GEO Company either; that would require its own question-and-answer sample.
The practical value is keeping two successful checks from being purchased repeatedly as though they explained the entire problem. Next work follows the missing evidence. It might examine recorded access, inspect a URL in the relevant search system or compare sources answering a commercial question.
Keep indexing and retrieval checks within their scope
For Google, examine information about the URL you control and distinguish the published version from the last version observed by its systems. The official recrawl-request guide explains the procedure and its limits. Requesting a crawl does not confirm immediate inclusion.
Google’s AI-features documentation connects supporting-link eligibility with indexing and snippet eligibility, without guaranteeing appearance. That condition belongs to those Google features. A Google URL status is not a universal explanation of every AI search product.
If important information is missing, compare the received source with visible content and identify its dependency: template, API, interaction or access. The technical GEO checklist develops those controls. Commission a correction to missing content before paying for an editorial strategy around a passage the request does not receive.
Assess whether the page answers the buyer
Read the sources in the observed answer and classify their contribution: definition, comparison, experience, requirements, location or evidence of a capability. Compare your page with that need. A service page can introduce the company beautifully while leaving its deliverables, audience and boundaries unexplained.
Review specific questions. Can a buyer identify the needed capability? Are conditions, examples and the next step available? Is the service boundary understandable? This produces a verifiable editorial brief and improves the source’s commercial usefulness, while source selection remains outside the supplier’s control.
Avoid rewriting the entire website around one answer. If a scope section closes the gap, commission that section. If the question expresses another intent, a new guide may be appropriate. Preserve the existing page’s purpose and connect the two with links that help the reader continue.
Review facts and evidence you can publish
A claim such as “industrial integration specialists” needs an explanation. Document the system, task, scope and publishable experience. Where a real case can be shared with permission, describe its context and measured results. Label examples and limit conclusions to what they demonstrate.
Check consistency between the website and relevant external sources. An old profile might describe work the business no longer offers. Correct facts through the available process and retain the reference. Invented reviews, mentions and supposed research create another information problem.
An AI search content strategy should begin with those questions and publishable evidence. It can connect a clear service page, a decision guide and a useful case. Publication volume follows the gap the programme needs to address.
Choose investment that matches the finding
| What is established | Initial work | Deliverable to request |
|---|---|---|
| Blocked response or missing content | Bounded technical diagnosis | Cause, proposed change and repeated check |
| Received page does not answer the buyer | Scope and content review | Approved passages and usefulness criteria |
| Sources contain contradictory facts | Information reconciliation | Approved source, corrections and owners |
| Useful source without a documented answer pattern | Visibility baseline | Questions, samples, sources and repeatable method |
Where several gaps exist, order the dependencies. Publishing a case before its facts are approved can create rework. Correcting a template that removes text can help several pages. Ask the proposal to explain that sequence and separate assessment, implementation and follow-up.
Build an answer sample that supports decisions
Choose discovery, comparison and purchasing questions relevant to your business. Keep stable identifiers and record intentional variations. Separate brand and category, markets and languages. The LLM visibility audit guide explains how to organise that baseline.
An absence is an observation under particular conditions. Repetition can reveal patterns when the method is retained. Changing questions, product and market together makes the earlier percentage less directly comparable. Annotate publications, corrections and campaigns that coincide with measurement.
Combine this review with attributable visits and commercial enquiries where data supports it. Keep mentions, citations and business results distinguishable. A cited link with no visits and a qualified enquiry need different interpretations; neither should vanish inside one overall score.
Questions to settle with a GEO supplier
What if we already have a technical audit?
Provide the original URLs, methods, dates and findings. Ask which checks remain valid and which need repeating because the page or infrastructure changed. A supplier should identify the additional evidence required for the current decision. A successful earlier check is useful context, even when it cannot establish today’s response.
Should we publish a new page for every unanswered question?
First map the question to a buying decision and existing source. Several questions can belong to one substantial guide; a separate commercial intent may justify another page. Require a useful contribution and an internal-linking role before commissioning each entry. Mechanical variations alone make a poor editorial brief.
Can the intervention be accepted without a new citation?
Yes, when its agreed deliverable is verifiable source improvement. A technical change can be checked on the published response; an editorial change can be reviewed against approved facts and buyer questions. Keep later visibility observation in the follow-up scope rather than treating a third-party citation as a promised deployment result.
A commissioning example: readable page, unclear scope
Consider a fictional provider whose integration page does not explain whether the service includes migrating a catalogue or maintaining it after launch. The commercial question asks about precisely those responsibilities. A first editorial engagement can document inputs, delivery, maintenance and exclusions with the service owner.
Acceptance reviews the approved facts and checks that the passage is published at the intended URL with a useful next step. Follow-up retains the original questions and records subsequent answers. This is not a customer case or evidence of citation improvement. It illustrates a bounded intervention the team can deliver.
If the review establishes that the provider does not offer the capability, change the commercial objective. Adding a keyword cannot create a service. That decision avoids spending on a page that attracts enquiries the business cannot fulfil.
Keep the next review useful
Agree what will be reviewed after implementation: publication integrity, access observations, a comparable answer sample and business enquiries where available. Assign each check to an owner and retain its date. A content lead, developer and commercial team may need different evidence from the same intervention.
Keep unresolved questions visible. If the answer sample still differs between markets, investigate that segment before expanding the entire content programme. If the source change was not published as approved, fix delivery first. The next action should follow the current observation rather than repeat the original package by default.
What to send for an assessment
Share the domain, priority URL, question, product used, language, market, date and archived answer. Include previous checks and their limits. A screenshot is a useful starting point; leave unknown fields explicit instead of filling them with assumptions.
For technical work, explore technical GEO audit and implementation. To compare deliverables, use the GEO audit report template for 2027. Request an assessment with that record: identify the decision needing evidence, then define the intervention. Your budget benefits when every piece of work has a concrete reason.