AI Trust AuditBETA
    Back to overview
    Product Updates

    Five new checks and a new point distribution in the AI Trust Audit

    GEO Expert14. September 20268 min read
    Five new checks and a new point distribution in the AI Trust Audit

    AI systems decide in seconds whether to use a website as a source. What they base this decision on has shifted noticeably in recent months: it is no longer just technical factors that decide, but whether statements are evidenced, answers are directly extractable, and information is consistent across all sources. Therefore, the AI Trust Audit now checks five additional points and evaluates existing criteria with new weighting.

    Key Takeaways

    • Five new checks: Evidence density, directly extractable answers, a real access test for AI crawlers, prices in the source code, and consistency of your core information.
    • New weighting: 17 criteria now count more, 29 less. What really matters for AI answers has more weight.
    • Everything remains comparable: No measurement results are overwritten. Older reports are displayed according to the current distribution, so that your progress over several audits is based on the same scale.

    Why the evaluation is evolving

    When the AI Trust Audit launched, websites primarily failed on basics: missing sitemaps, blocked crawlers, no data markup. These fundamentals are now in place for many websites. What makes the difference today is one level above: a language model prefers to cite content that it can adopt without interpretation and that can be verified against other sources.

    This is exactly where the new checks come in. Each of them answers a question that an AI system implicitly asks before it mentions your website in an answer.

    The five new checks

    Evidence density in content

    This checks how many verifiable numbers, dates, sources, and quotes your content contains. A page that promises "market-leading quality" provides a language model with nothing citable. A page that writes "on the market since 2008, over 40,000 orders delivered in 2025, certified according to DIN EN ISO 9001", does.

    What you can do: Add specific numbers, dates, and source citations to your most important pages. One to two verifiable facts per section are usually enough to make the difference.

    Directly extractable answers

    Here we check whether question headings are present with a short, complete answer directly below them. Such blocks are ideal for AI answers because they can be adopted without summarization.

    What you can do: Phrase headings as actual customer questions and answer them in the first sentence below—briefly, independently understandable, without referring back to the previous paragraph. The detailed explanation can follow after that.

    Access for AI crawlers (Live test)

    Previously, we evaluated what your robots.txt allows. Now we also check whether your page is actually accessible using the real identifiers of GPTBot, ClaudeBot, PerplexityBot, and Google-Extended.

    This is an important difference: many websites allow AI crawlers on paper but block them in practice via a security solution, a firewall, or bot detection in hosting. This case often goes unnoticed for years because everything works in a normal browser.

    What you can do: If the check triggers an alert, speak to your hosting or security provider and specifically exclude the mentioned identifiers from the blockade.

    Prices and availability in HTML

    This checks whether price and availability are already in the delivered source code and not just reloaded via a script. For visitors, this makes no difference. For a language model that reads the page without script execution, it determines whether your offer is captured with a price at all.

    What you can do: Have the price and availability output server-side and additionally marked up in the data markup. Reloaded versions for discounts or tiered pricing can remain as long as a base value is in the source code.

    Consistency of core information

    Company name, address, and phone number are compared between the homepage, legal notice (Impressum), and data markup. Discrepancies almost always happen unintentionally—a move, a new phone number, a legal form updated in one place but not in three others.

    For AI systems, such contradictions are a trust issue: if your own information contradicts itself, the probability that your brand will be recognized and mentioned as a unique entity decreases.

    What you can do: Define a mandatory spelling for name, address, and phone number and reconcile the legal notice, contact page, footer, and data markup completely once.

    The new point distribution

    Parallel to the new checks, we have revised the weighting of the entire catalog. 17 criteria now count more, 29 less, and 29 remained unchanged. The total achievable score thus decreases from 170 to 153.

    The direction behind this: criteria that actually decide whether content is cited carry more weight. This includes structured content formats such as tables and numbered lists, the linking of your brand with authoritative external profiles, and the quality of your about-us page. Criteria with less influence on actual visibility were reduced in return.

    Do my previous results remain comparable? Yes, and we solved it this way intentionally: your measured individual results remain stored unchanged; nothing is overwritten retroactively. However, all reports—including older ones—are displayed according to the current point distribution. This is the only way two audits of the same website are based on the same scale, and the progression says something about your website instead of the evaluation version. In the report, you will also see the previous value for comparison.

    A note for context: a slightly changed result after the transition does not mean that your website has become better or worse. It means that the evaluation more accurately reflects what AI systems pay attention to today.

    Your next steps

    1. Start a new audit of your website so that the five new criteria and the current weighting are included.
    2. Look at the access test for AI crawlers first. An unnoticed blockade costs visibility regardless of how good your content is.
    3. Next, take your three most important pages and add verifiable facts and clear question-answer blocks.
    4. Reconcile your core information across all pages once.
    5. Audit again after the adjustments and compare the percentage value with your previous result.

    The criteria continue to evolve because the systems that decide your visibility also continue to evolve. We document every change transparently so that you can correctly categorize your results at any time.