Google Validated the Fundamentals for AI Search. We Already Check Them All.

In May 2026, Google published its first official guide to AI Overviews and AI Mode. Its central argument: good SEO is enough. No special tactics needed. We agree, for Google. But ChatGPT, Perplexity, Claude, Grok, and DeepSeek work differently, and that is where the AI-specific signals still matter. Here is what we changed and why your score might have shifted.

What Google Actually Said

Google's guide, Optimizing your website for generative AI features on Google Search, was published in May 2026. It is the first time Google has published dedicated guidance on AI Overviews and AI Mode.

The guide's thesis is straightforward: SEO best practices continue to be relevant because Google's AI features are rooted in its core Search ranking and quality systems. It cites RAG (retrieval-augmented generation) and query fan-out as the techniques connecting SEO fundamentals to AI responses. Good content, clean technical structure, structured data, crawlability, and page experience. Everything we had already been checking.

The guide also has a "mythbusting" section that tells site owners what they don't need to do, including creating "new machine-readable files" or using special AI-specific schema. Google is being honest: for Google's AI features, standard SEO is sufficient.

That is fair. It is also only half the picture.

What Google Got Right (and We Already Check)

The guide's recommendations line up with what we knew from testing and from the Google API leak. Good SEO fundamentals matter for AI visibility. We had already been checking most of these signals. The guide confirmed it, and we filled the gaps.

What the Guide Does Not Cover

Google's guide speaks for Google. AI Overviews and AI Mode use Google's existing index and standard schema types, so standard SEO is genuinely enough for those features.

But most people using AI search in 2026 are not only using Google. ChatGPT, Perplexity, Claude, Grok, and DeepSeek each crawl the open web independently and respond to signals Google says you can skip:

• llms.txt - a proposed standard that some AI platforms may use to discover site content. Google explicitly says it is not needed for Google AI features. We still check it because it may help with non-Google platforms, but we no longer weight it as heavily.
• AI bot access - PerplexityBot, ClaudeBot, and GPTBot each need explicit permission in robots.txt. Google's crawler permissions are separate.
• Citable passage structure - Google's guide says write well for humans and you are done. In practice, ChatGPT and Perplexity tend to extract self-contained 100-200 word paragraphs more reliably than loosely structured prose.

This is the gap Google's guide does not address. SearchScore exists precisely because visibility across six AI engines requires more than just good SEO for Google. The audit runs Google-aligned checks and non-Google checks side by side, and the fix list now labels which signals help which platform.

What Changed in the Audit

12 new checks aligned with Google's recommendations

These new checks test signals the guide recommends or implies:

• VideoObject schema (Structured Data, 8 points) - the guide says video can appear in AI responses; we now check for proper VideoObject JSON-LD
• Image dimensions (Technical, 3 points) - layout stability matters for page experience
• Responsive images (Technical, 2 points) - srcset and picture elements for device-appropriate delivery
• Descriptive image filenames (Technical, 2 points) - flags generic names like img001.jpg
• Domain consistency (Technical, 3 points) - www/non-www redirect to a single canonical
• Trailing slash consistency (Technical, 2 points) - mixed slashes create duplicate URLs
• Parameter deduplication (Technical, 2 points) - tracking parameters in canonicals
• Unique POV signals (E-E-A-T, 4 points) - the guide specifically values "unique point of view" and "non-commodity content"
• Image captions (E-E-A-T, 2 points) - the guide mentions supporting text content with images
• Gzip/Brotli compression (Technical, 3 points) - page experience signal
• Long-term cache headers (Technical, 3 points) - page speed and experience
• Resource hints (Technical, 4 points) - preconnect, preload, dns-prefetch

llms.txt re-weighted, not removed

Previously, llms.txt carried 27 points across two categories in the GEO score. Google explicitly says it is not needed. We reduced it to 14 points, reflecting its status as a proposed standard rather than a confirmed requirement.

The check is still there. It still fires. It still appears in the fix list for sites that want to maximise their chances across all platforms. But the fix text now clearly states: "Google does not use llms.txt - it relies on standard crawling and structured data." The reader can decide whether to implement it based on whether they care about non-Google platforms, not just Google.

Source attributions on every fix

Fix recommendations now include source attributions where they map to Google's published guidance. For example, the unique POV fix says: "Google's AI guide specifically values unique point of view and non-commodity content." Fixes for signals Google does not reference are labelled clearly as non-Google signals.

Why Your Score Might Have Changed

If you ran an audit before this update and run one now, your GEO score may have shifted:

• Score went up: Your site was already doing things the guide recommends (video schema, image dimensions, unique content) that we were not previously scoring. Those now earn points.
• Score went down: You were earning points from llms.txt (which Google ignores) but missing checks we have now added. The score now reflects a broader and more honest picture across all six engines.
• Score stayed the same: Your site was already strong on both Google-aligned and non-Google signals.

The updated score reflects visibility across six AI engines, each with different requirements, not just Google.

Check Your Site

The updated audit is live now. Run it on any domain and you will see the new checks, source attributions, and clear labelling of which signals help which engine.

The Full Methodology

The complete list of all 250+ signals, scoring weights, and category breakdowns is documented on our methodology page. The Google alignment section explains exactly how each audit category maps to Google's published recommendations.