Which Tools Do We Use for AI Visibility? Our Tool Set and When We Don't Use It

Correction (6 October 2026): Our measurement provider withdrew two data sets used in this post. The 5 October 2026 scan was only partially measured: the provider ran out of credit during the scan; ChatGPT answered all 300 questions, while Gemini, Perplexity, Claude and Grok answered only 42–44. The Gemini and Google AI Overview zeros in the 15 September 2026 scan were also a measurement error. Those numbers in this post are invalid. The 15 September ChatGPT measurement (249 answers, 11.2%, ±3.9 points) stands. We will update this post when a complete new scan is available.

Short answer: To measure AI visibility we use third-party platform scans and a second measurement line where we document the model version ourselves. For organic search and traffic we look at Google Search Console and GA4. For technical audits we use Semrush, and before anything goes live we rely on our own build checks. No single tool's output counts as a decision. Each tool answers one question; the decision comes from reading several of them side by side.

In this post we explain why we use each tool and when we do not. For a general comparison of third-party tools, see our AI visibility and reputation tools guide.

Measurement: do we appear in AI answers?

Third-party measurement platform

What we use it for: Mention rate per engine, margin of error, competitor share of voice and the cited URLs. Our measurement infrastructure is a third-party platform.

When we do not use it:

• To compare the overall scores of two scans with different engine sets. The engine set can change from scan to scan; different engines ran in our September 15, 2026 and October 5, 2026 scans.
• To apply automatic strategy suggestions as they are. We first match each suggestion against the existing page on the site and against our own data. For example, the October 5, 2026 report suggested rewriting one of our guides; we kept that page and covered the topic in a separate post.

Second measurement line

What we use it for: Recording the model version, whether web search was on, a timestamp and the full response text for every response. On October 5, 2026 we collected 32 responses from 4 models on this line.

When we do not use it: To report rates. Its sample is small; it shows a direction, not a statistical result. We report rates from platform scans, with margins of error.

Organic search and traffic

Google Search Console

What we use it for: Which page is shown for which query, position and clicks. We built our internal linking plan on GSC data.

When we do not use it: To measure AI visibility. Ranking high on Google does not mean being mentioned in an AI answer. We showed this with our own data in our zero-click search post.

GA4

What we use it for: Seeing the volume and quality of visits from AI assistants: sessions and time on site.

When we do not use it: To count how often AI recommended us. If a user sees the recommendation and later comes to the site directly, GA4 records it as "direct". The recommendation stays inside the conversation.

Technical audit

Semrush site audit

What we use it for: Finding crawlability, schema and hreflang errors in bulk. On September 26, 2026 we fixed the audit findings; we brought wrong hreflang tags from 54 to 0 and invalid schema properties to zero across 189 pages.

When we do not use it: To apply a finding without checking it. We check every finding against the schema.org vocabulary and the page's actual HTML.

An open-source GEO analysis tool

What we use it for: A quick scan of a page's openness to AI crawlers and its content structure.

When we do not use it: For the schema score. We found that the tool we use does not read pages that deliver schema in a single @graph block correctly. Our site delivers schema that way. So we do not use the tool's schema score in our reports.

Before publishing: our own build checks

Before every release the site's build step runs three checks:

1. Schema validation. Parses the JSON-LD blocks on every page and lists errors and warnings.
2. Link validation. Lists any broken internal link.
3. Unsourced-figure check. Flags percentages and rates that have no source, date or margin of error next to them. It also catches invented institution and method names.

We added the third check on September 16, 2026, after finding units cited as sources on the site with no measurement behind them.

An example: one question, four tools

The platform's citation list from the October 5, 2026 scan showed that 13 of 25 citations went to our English /en/ pages, although most queries were in Turkish. On its own this is an observation. To turn it into a decision we look at the other tools:

1. Search Console: Is the Turkish page shown on Google for Turkish queries?
2. Semrush audit: Is the hreflang mapping correct? Have the errors we fixed on September 26 come back?
3. Build checks: Are the Turkish page's schema and links clean?
4. The next scan: After the fix, did the citation move to the Turkish page? We only look at the same engines.

No tool answers this question on its own.

Conclusion

Choosing tools is not a list; it is matching tools to questions. Measurement tools answer "do we appear in AI answers?". Search Console answers "where do we stand on Google?". Audit tools and our own checks answer "is our site technically readable?". Knowing what a tool does not measure matters as much as knowing what it does.