Methodology: exactly what the Venoh diagnostic checks (and what it does not)

Step 1: fetch real pages

The diagnostic starts from your homepage. It performs a normal web request, follows ordinary redirects, and reads the HTML that comes back. If the homepage turns out to be an almost empty stub whose only job is a simple browser-side redirect (for example, a language-selection page that sends visitors to a language version), it reads the destination written in the page and analyses that page instead, on the same site only.

It then looks at the homepage's own navigation (links in the header, navigation and footer areas) and selects up to four more pages that matter most for AI answers: typically FAQ, services or products, pricing, about, and case-study pages, whichever the site actually links to. It never follows links to other websites. Before each additional page it reads your robots.txt and skips any path disallowed for all crawlers (User-agent: *).

Requests identify themselves honestly with the user-agent venoh-geo-research-bot/0.1 (+https://venoh.com). The diagnostic uses public pages only: no logins, no CMS access and no code changes on your side.

Step 2: extract concrete signals

From each fetched page the tooling extracts the following without any interpretation, so each item can be pointed to in the page source:

• The <title> and meta description.
• Headings (H1 to H3).
• The visible text and its word count.
• Structured data: schema.org types declared in JSON-LD (including nested items and @graph) and in microdata.
• Whether FAQ-style content is present, from the headings and text or from FAQPage markup.
• Same-site navigation links, used to choose the extra pages in step 1.

Step 3: ask AI answer engines your buyers' questions

For the paid diagnostic, we work out around 30 questions a buyer in your category would ask (they never mention your company, so the test is not biased toward you) and put each one fresh to AI answer engines with live web search switched on. For every answer we record which sites were cited, whether yours was among them, and which other companies' sites appeared instead. The counts in the report are tallied by code from those answers, not written by a model.

The report names exactly which engines were tested and on what date. AI answers vary from day to day and from one wording to the next, so this is a dated sample, not a ranking or a promise about tomorrow's answers.

Step 4: write up findings with an evidence label

The extracted signals are analysed and written into findings. Every finding carries one of three labels so you can see how much weight it deserves:

If a page cannot be fetched or read, the diagnostic says so and marks the affected areas NOT_CHECKED. It does not fill the gap with assumptions. A finding that rests on a page the tooling could not read is a defect in the diagnostic, not a fact about your company, and we treat it that way.

Step 5: prioritize, then check the write-up

Findings and recommendations are each given a priority (high, medium or low) so you can see what to fix first. The report is written to be implementation-ready: a developer or content lead should be able to act on each item without a follow-up meeting to decode it.

Before the report goes out, a separate model reads every statement in it against the captured page evidence and the engine results. Any statement it cannot back up stops automatic delivery until a person has read the report.

What the diagnostic does not do

Being precise about limits is part of the method. The diagnostic is a preliminary pass. It does not:

• Audit your whole site. It reads the homepage plus up to four relevant pages.
• Analyse backlinks, domain authority or off-site mentions.
• Track AI answers over time. The engine test is one dated snapshot, on the engines the report names; engines it does not name were not tested.
• Run JavaScript. It reads the HTML a crawler receives; apart from following simple, literally written redirects, content that only appears after scripts execute is not seen (which is itself often a finding).
• Measure page speed, Core Web Vitals or accessibility.
• Review content you have not published, or anything behind a login.
• Guarantee that any change will lead to a citation or a ranking. No one can.

What you receive

• A written report of evidence-labelled findings, each tied to the page it came from.
• A prioritized list of recommendations, written to be actioned directly.
• A clear list of what was not checked, so nothing is implied that was not examined.

For the deliverable and pricing, see pricing. To understand the concepts behind the checks, read What is GEO? and the how to get cited by AI checklist. Terms are defined in the glossary.

Opting out or asking questions

If you would rather the diagnostic did not fetch your site, or you have seen the Venoh user-agent in your logs and want to know why, write to us through the contact form and we will respond. Besides paid diagnostics, we run a lighter version of these checks on the public pages of companies we may contact about AI search visibility. Either way it makes a small number of ordinary page requests per run, and if you ask us not to, we will not fetch your site again.