The GEO industry just got graded.

Google updated three Search Central documents this week, and the combined effect is something the search industry has not had before: a short, public rubric for evaluating GEO agencies and AI search tools. Cut the marketing language out of the new pages and what's left is a five-question scorecard any buyer can hold up to any vendor pitch.

Most of the industry agencies and tools both will not pass it cleanly.

Paired with the most rigorous study of fan-out optimization ever published, the picture this week is clear: the tactics being sold loudest are the ones with the weakest evidence underneath them. The teams building defensible AI visibility right now are doing something quieter, more operator-led, and a lot harder to package.

Here's what we're tracking.

The most consequential line from Google's new third-party SEO tools page: "Third-party tools don't have access to our internal ranking data." Anything a tool reports about your visibility is its own observation, not a window into Google's systems.

The rubric, distilled from all three updated pages:

Any vendor that answers cleanly on all five is operating inside Google's stated terms. Any vendor that flinches on two or more is selling something Google has now explicitly cautioned against.

The companion AI optimization guide names three tactics directly: llms.txt files (Google does not use them), artificial content chunking (Google's systems handle multi-topic pages without help), and pursuit of inauthentic third-party mentions (its anti-spam systems already act on this). The headlines framed this as Google "condemning" GEO. The actual text is more measured Google explicitly acknowledges third-party tools can be useful. The criticism is precise: it targets vendors invoking Google's name to imply endorsement, vendors selling predictions as Google data, and vendors guaranteeing outcomes nobody can guarantee.

A vendor who welcomes these questions on a sales call is the one Google's guidance was written to protect. A vendor who deflects is the one it was written about.

If you've been pitched a "query fan-out optimization" service in the last six months, this is the data your team needs.

The headline finding, in Indig's own words: fan-out coverage is nearly irrelevant to citation rates.

The five things worth internalizing:

The mechanism is real and triangulated. Fan-out is documented in Google's own product announcements, in three Google patents, and in OpenAI's API behavior. The optimization paradigm built on top of it — the tools, dashboards, "fan-out coverage" KPIs — is mostly buzzword. Run those tools through Google's new five-question rubric and most score badly on question 3 (observed tracking vs. claimed signal access) and question 5 (advice consistent with Google's guidance).

While the GEO industry argues about whose dashboard is most accurate, a parallel motion is happening among in-house operators. Adam Robinson (Retention.com / RB2B) recently shared how his team used Clay, Claude Code, and Apify together to ship a launch — assembling research, enrichment, and execution inside an agentic stack rather than buying it from any one platform.

The specific tooling matters less than the pattern. Clay handles structured data and waterfalls. Claude Code handles logic, content, and orchestration. Apify handles the scraping and retrieval layer. The combination replaces what used to require three different vendor contracts and a quarterly QBR.

Why this matters for GEO specifically: the same stack — agentic orchestration plus enrichment plus retrieval — is what makes brand-mention monitoring, citation auditing, and proprietary-data publishing cheap enough to do continuously rather than quarterly. It's the operational version of last week's Citation Paradox argument: original information becomes a moat only if you can produce it at speed.

The teams that compound through 2026 won't be the ones with the most vendors. They'll be the ones whose internal team can build, test, and ship a new measurement view in a week — and pass Google's five-question rubric on every claim they make to their executives about what the numbers mean.

Google's rubric draws a sharp line between claims of Google data and honest observed tracking. The Brand Mentions view inside Passionfruit Labs is built to sit cleanly on the observed-tracking side.

A snapshot from a recent workspace (Client) across May 10 – June 9:

The Recent Conversations table is where the operational value lives. For Advaiya, every cited mention names a specific product — Order Short Close extension, HSE Score Tracker, InterestCalc Pro, ProcureSure, AdLead, Insights Central, Project Changes Tracker. AI isn't citing them for "what is a Dynamics 365 ISV." It's citing them for named proprietary products only they ship.

That's the Citation Paradox confirmed in workspace data. The brands earning the citations that compound are the ones publishing things models can't generate on their own — named products, specific use cases, proprietary integrations. Track that pattern at the conversation level, and you know which assets are doing the citation work and which ones are flattering vanity dashboards.

One question to audit: For the 5 non-branded queries that bring you the newest buyers, what's your share of voice across ChatGPT, Gemini, and Perplexity? If you only know your branded numbers, you're tracking the comfortable half.

One quick win: Publish one 10–20-minute video review in your core product category this week. Title it around what you tested and what you found — not "tutorial" or "explained." That single asset can capture the video citation slot for an entire query category on Gemini.

One thing to stop doing: Building "What is X" content as your primary AI-search play. Google's own usage data shows planning and decision queries growing 80% faster. Rebalance toward comparison and decision content.