Updated 16 October 2026. This article replaces unsupported agency allegations, an unverified events-business case story and universal claims about AI crawling and result timelines. The guidance below distinguishes technical findings from observed assistant answers.
An absent recommendation is a starting point for an investigation. It is a poor diagnosis on its own. The assistant may have interpreted the question differently, selected other sources or answered without a fresh search. There may also be a real problem with the website. You need evidence to tell those possibilities apart.
For a recruitment owner, the question is whether an employer can find the relevant service, understand the agency’s experience and start a conversation. Test that before declaring the business invisible.
Capture the answer before changing the site
Save the exact question, the product and mode, the date, any relevant location setting, the answer and its cited URLs. Record whether the session was signed in or carried previous conversation context. If the product was inaccessible or the request failed, record the failure separately from an answer that omitted the business.
Keep a small set of realistic buyer questions fixed for later comparison. Test the named agency as well as an unbranded supplier question. Avoid changing the wording repeatedly until a preferred result appears; the original evidence still needs explaining.
Compare the first response with the finished page
Open the relevant service page and inspect the HTML returned by its URL. Then compare it with the rendered browser page. Look for the important facts: company identity, service scope, roles covered, location and a usable enquiry link.
If those facts appear only after JavaScript runs, a retrieval tool that does not render scripts may miss them. That finding can justify a rendering repair. It does not establish that every AI system sees a blank page. Google Search can render JavaScript, and browser agents may use the rendered screen, page structure and accessibility information.
Check the actual delivery before blaming the platform name. Two sites built with the same framework can expose their content differently. A full rebuild should answer an observed problem that a smaller change cannot reasonably solve.
Check the relevant access controls
Read the robots rules for the actual user agent and URL. Inspect response status codes and any available verified crawler or firewall records. A successful request from your own laptop does not prove what a service’s crawler received.
Separate search crawling, potential model training and user-requested fetching. OpenAI, Anthropic and Perplexity document distinct roles. Blocking a training crawler does not automatically prove absence from every answer route; allowing one does not guarantee inclusion. Keep private account pages and reports private while investigating public discovery.
Look for an information problem
Ask someone unfamiliar with the agency to identify its specialism, the businesses it serves and how to enquire. If the page requires several guesses, an assistant may also produce an incomplete description. Fix the ambiguous wording because it matters to the buyer.
Check for conflicting company names, expired offers, unsupported testimonials and service claims that differ across pages. Make internal links useful and descriptive. A heading called “Solutions” followed by another heading called “Excellence” can leave everybody fully informed about absolutely nothing.
You can supply the page to an assistant and ask factual questions as a separate extraction test. Label that test clearly. Success with a supplied URL does not demonstrate unaided discovery.
Validate schema without making it a gatekeeper
Structured data should describe supported facts visible on the page. Validate it and inspect the relationships between the organisation, service and article. Do not add reviews, offices or qualifications merely to make the graph look more substantial.
Google’s current guidance says structured data is not a requirement for its generative search features. The Rich Results Test covers supported Google features, while a general schema validator examines the wider vocabulary. Passing either check does not certify that an assistant will cite the business.
Treat llms.txt as an optional maintained resource for tools that choose to use it. It is not an access-control file, and Google Search says it does not use it to improve or reduce search visibility.
Measure performance with its limits attached
Investigate slow or failed responses, broken navigation and poor mobile usability. Record the URL, device conditions, date and test method. A lab performance score and real-user field measurements answer different questions, and some pages have insufficient field data.
There is no verified universal 200-millisecond abandonment rule for every AI crawler in the sources used here. Improve faults that the measurements demonstrate. Do not turn a fast test result into a claim about citations, or assume that an assistant’s absence proves the hosting is too slow.
Decide what happens next
Write a repair list with the evidence beside each item, then repeat the same technical checks after publication. Revisit the fixed buyer questions on an agreed schedule. Keep mentions, accurate descriptions, citations, visits and enquiries separate so a technical improvement cannot accidentally become a claimed sales result.
ScopeSite’s published H4TLT case study provides an approved example from workplace hearing surveillance. Its evidence belongs to that work and should be read directly, without borrowing its result for another sector.
CAFMO is live at canaifindme.online. Can AI Find Me Online is ScopeSite’s application for page checks, model answer evidence and a saved improvement plan, with monthly subscription plans. V.O.I.C.E. remains the methodology and engine name. The earlier scanner keeps separate accounts and credits; purchased legacy credits do not expire or automatically become a subscription. Model API observations can differ from the consumer assistant applications.
Read the current scanner status or the recruitment discovery guide, then start with the question and page that matter to the next employer.



