When a tool gives a website an AI visibility score, the first question should be what it measured. Did it inspect page features, observe a citation in an answer or record a visitor who became a customer? Those are three different results. Auriti Labs’ GEO Optimizer is useful to examine because its documentation exposes the scoring rules and also describes separate citation checks.
The toolkit can run through a command line, Python and MCP. It checks areas including crawler access, metadata, structured data, content and identity signals. Its audit score uses the project’s own weights. A separate citation command samples supported providers, with important differences between providers that expose source URLs and those that only reveal model responses.
I would use those outputs as evidence for specific investigations. I would not translate a score of 90 into a claim that a company will appear in 90 percent of AI answers.
Separate readiness, observation and business value
A readiness check can reveal a missing canonical or content that is difficult to retrieve. A citation observation can show that one answer used your page. A qualified enquiry can show that somebody found your offer relevant. Improving one layer may help the next, but they should remain separate in a report.
For a service website, my proposed baseline would include a small set of genuine customer questions. Mix an explanation question, a supplier-selection question and a problem-specific question. Keep language and market explicit. Repeat the same set after changes, record the provider and date, and retain the actual source URLs when available.
This avoids a common measurement trap: changing the prompt until the business finally appears, then calling that a visibility improvement. The question set should be decided before the results are examined.
Read the rubric before assigning development work
GEO Optimizer gives weight to files and signals such as llms.txt. That makes their presence relevant to its score. It does not establish that every search or answer engine requires those files, or that adding one creates citations. The tool’s own estimated score after a fix is also not an observed engine outcome.
| Audit output | How I would use it |
|---|---|
| Content cannot be fetched. | Reproduce the response and determine which access path is affected. |
| A schema recommendation is missing. | Check whether that schema accurately describes the visible page. |
| A suggested discovery file is absent. | Assess its purpose separately from the points it adds. |
| A sampled answer cites a competitor. | Inspect the actual cited page and the question it answers. |
Prioritise changes that also help a visitor understand and trust the business. Clear service scope, maintained product facts and an accessible page have a defensible purpose even when an answer engine’s behaviour changes.
Use a controlled before-and-after comparison
Suppose a company has an unclear implementation page. I would first save its current text and measure the agreed question set. Then improve the explanation of the customer problem, deliverables and relevant experience. Record what changed before repeating the checks. This is a proposed evaluation method, not a claim about a tested client result.
If citations change, examine the cited URLs and the answers themselves. If nothing changes, the page may still be more useful to visitors. Neither result justifies inventing a causal percentage from a handful of responses. Small samples should remain small samples in the report.
For first-party search evidence, mcp-gsc provides a different view. For websites that want agents to perform tasks after discovery, WebMCP addresses an interaction problem. An accessible tool interface and a cited editorial page are different deliverables.
Choose an improvement your team can maintain
My preferred audit ends with a short work list, evidence for each item and a retest method. It should say who maintains service facts and how outdated claims are corrected. That creates a useful operating habit rather than a one-time chase for a perfect badge.
If you want to connect AI-search checks with crawlability, content quality and enquiry measurement, my technical SEO and measurement work can help define and implement that scope.
Source checked 7 October 2026: the GEO Optimizer repository README and its published scoring description. No tool score, citation gain or customer outcome was measured for this article.