What Google's first AI optimisation guide gets right, and what it leaves out

In partnership with

AI Agents Are Reading Your Docs. Are You Ready?

Last month, 48% of visitors to documentation sites across Mintlify were AI agents, not humans.

Claude Code, Cursor, and other coding agents are becoming the actual customers reading your docs. And they read everything.

This changes what good documentation means. Humans skim and forgive gaps. Agents methodically check every endpoint, read every guide, and compare you against alternatives with zero fatigue.

Your docs aren't just helping users anymore. They're your product's first interview with the machines deciding whether to recommend you.

That means: clear schema markup so agents can parse your content, real benchmarks instead of marketing fluff, open endpoints agents can actually test, and honest comparisons that emphasize strengths without hype.

Mintlify powers documentation for over 20,000 companies, reaching 100M+ people every year. We just raised a $45M Series B led by @a16z and @SalesforceVC to build the knowledge layer for the agent era.

Make Your Docs Agent-Ready

🤖 Google publishes its AI optimisation guide

Useful for Google Search. But remember, it's only Google's view of the discipline.

This week Google dropped its first official guide to optimising for generative AI features in Search. For a company that has shaped 25 years of SEO, this is a big moment. We've had John Mueller, Gary Illyes and other spokespeople drip-feeding hints and insights for the last three years, but never a single, named document like this.

The guide explains how AI Overviews and AI Mode work. Two mechanics matter:

Retrieval-augmented generation (RAG) — the system retrieves live information from Google's index to ground its answer. Also called grounding.
Query fan-out — your prompt gets broken into multiple sub-prompts, and the model retrieves against each one. An example of a Query fan-out taken from Grok using Obsero is below.

web_search_query: [
  "best AEO platforms for tracking AI search visibility",
  "AEO tools comparison AI search tracking 2025 2026",
  "top Answer Engine Optimization platforms",
  "\"AEO\" OR \"Answer Engine Optimization\" tracking tool OR platform"
]

Understanding fan-out matters. These sub-prompts are probabilistic (they change every time) so capturing a single snapshot is not enough. But if you track them over time, you can start to see directionally what the AI is asking on your behalf. That tells you what topics, entities and questions the AI considers adjacent to a given prompt, which is where the actual optimisation work lives.

What Google gets right

The bulk of the advice is solid and unsurprising. Create non-commodity content. Provide a unique point of view. Use high-quality images and video. Don't write walls of text. Make sure AI-generated content meets the spam policies. Build a clear technical structure.

This is foundational SEO with the labels swapped. And foundational SEO is still doing a lot of work in this new world. If your technical foundations are broken, you are not winning anywhere - RAG-grounded answers included.

The point about a unique point of view is the one I'd underline. Storytelling, brand narrative and your own first-party data are how you separate yourself from the commodity layer. If your content could have been written by anyone, it probably was - and the model will treat it that way.

What Google conveniently leaves out

Here's where I'd push back. This guide is Google's perspective on Google Search. It is not a guide to AI search. Those are different conversations.

Google can only speak for Google. The guide explains how to optimise for AI Overviews and AI Mode, both of which run on Google's index. It does not explain how to show up in ChatGPT, Claude or Perplexity, because Google has no commercial interest in helping you do that. The advice is calibrated for one platform out of five. Treat it accordingly.

The "myths" section is the most-discussed part of the guide. My read:

LLMs.txt files: I don't recommend them for clients today, but never dismiss something for future use. Things change.
Chunking content: This one confuses what AI systems do with how you should write. If you take a wall of text and add headings, paragraphs and structure, you get semantically clean HTML. That isn't chunking. That's writing well.
Rewriting content for AI systems: Again, it's interpretation. "The company grew really big because of their growth plan" is not useful or descriptive. "Bob's Cola grew sales 27% in 2025 after a £4m billboard campaign across London transport, proving physical placements still convert in a digital-first market" is. That isn't writing for AI. That's writing properly.
Inauthentic mentions: Crudely fake mentions don't move the needle, fine. But LLMs build their picture of a brand from everywhere: blogs, forums, podcast transcripts, YouTube comments, listicles. Earned mentions across these surfaces genuinely do feed AI visibility. Google polices what it can rank-demote. It can't rank-demote a Reddit thread or a YouTube transcript, so of course it's telling you they don't count. Make of that what you will.

Worth remembering: Google spent years telling us clicks weren't a ranking signal, Chrome data wasn't used in Search, and site-wide authority didn't exist. The DOJ trial confirmed all three are wrong.

The bigger issue is what the guide skips entirely: how AI systems weigh parametric knowledge - the information baked into the model during training - versus what they retrieve live. RAG is one layer. Training data is the other. Google's guide only talks about one of them. Building an authoritative brand is key here.

The frame to take into the boardroom

If your CEO wants to talk about GEO, or AEO, or whatever the next acronym is, don't tell them “it's just SEO”. It isn't. The success metrics have changed. We are now measuring how often a brand is mentioned, the sentiment of those mentions, and whether they translate to business outcomes. Customer behaviour is fragmenting. Google Instant trained us to type short queries. Prompts in ChatGPT or Claude are longer, more personal, more detailed. The game has changed even if the SEO foundations underneath haven't.

Test, learn, never assume. That is your responsibility. That's the job.

📰 Condé Nast tells teams to plan as if search is zero

Large brands and loyal niches are holding. The middle is sagging.

Roger Lynch, the Condé Nast CEO, told his teams to plan their businesses as if search traffic will be zero.

Not literally zero. He expects it to settle at single-digit percentages of total traffic. But zero as a planning assumption. Why? Three years running, Condé Nast's internal forecasts predicted search traffic decline, and every year the actual decline was worse than forecast. So he told teams to stop forecasting and plan as if it's gone.

His comparison was sharp. Seven or eight years ago, a search showed a few sponsored links and ten blue links. Today you get an AI Overview, then rows of e-commerce, then sponsored stuff. He has to scroll to page two for an organic result.

This isn't just about AI search. It's about how Google has progressively monetised page one. The old paid-versus-organic split has merged into a single commercial proposition.

The barbell

Lynch's most useful observation is what he calls the barbell effect. Two ends of the publisher market are doing fine. The middle is dying.

Large authoritative brands — Vogue has grown revenue and profitability every year he's been there. The New Yorker had its best year ever.
Small niche brands with loyal audiences — Pitchfork is about 1% of revenue but has a defensible, paying niche.
The middle — broad, generalist publishers without deep authority or a specific niche have no clear path.

This isn't isolated. Chartbeat data shows search referrals down 60% for small publishers over two years. A Reuters Institute survey found media leaders expect search traffic to fall 40%+ over three years. Meanwhile Liz Reid, Google's VP of Search, has framed the losses as Google removing "low-quality bounce clicks", without sharing the data to back that up.

What to do about it

Condé Nast has doubled down on subscriptions. Digital subs were up 29% last year, prices have been raised materially, and retention improved every year rather than dropped. Pitchfork and Tatler have both launched paid subscriptions recently.

The lesson for any publisher relying on open-web crawl and ad revenue: you're exposed. If what you publish is worth paying for, charge for it. If it's worth crawling, monetise the crawl. The open-web-plus-ads model is the worst place to be sitting right now.

The lesson for brands: the same barbell applies to you. Large category leaders and tightly focused niche specialists are surviving the shift. Generic mid-market brands without deep authority or a defensible niche are the most exposed.

🛠️ Obsero updates

The new LLM breakdown view. Every platform, every competitor, one screen.

Three things worth sharing this week.

1. New trial journey

We've relaunched our trial flow this week using a mix of Claude Design and our amazing engineering team. The goal: help prospects get a sense of how Obsero works, see the social proof, and decide quickly whether it's for them.

Key changes:

50 free prompts at sign-up, up from 20
No credit card required — unlike many of our higher-profile competitors
Designed for small and medium businesses, not just enterprise

We're deliberately not chasing enterprise-only deals. If AI search visibility matters to your business, it shouldn't be gated behind a six-figure annual contract.

Step 1 of a simple signup process that gives you 50 free prompts to track in Obsero across Google AIOs and ChatGPT.

2. New client signed

This week we signed a major UK property brand. They've been in our pipeline for some time, are fully focused on AI search growth, and scored 8.3 out of 10 on our ICP calculator before we'd even started the conversation.

The ICP work is something I'd push every founder to do. If you're talking to everyone, you're talking to no one. Score your prospects against a clear ideal profile and the conversations get sharper, the proposals get sharper, and the close rates follow.

3. LLM breakdown report

We've also launched an LLM breakdown view, so you can quickly see how you're performing across ChatGPT, Claude, Perplexity, Gemini and Google AI Mode side by side. If you want a walkthrough, reach out for a demo.

That’s it for this week. Until next time.

Andy

ICYMI: three conversations worth your time

If you're producing content for AI search, working in news SEO, or trying to figure out how to actually report on visibility, these three are the ones to start with.