What Google's AI Optimization Guide Says About AI Mode
Sameer runs marketing for a pump manufacturer in Coimbatore. Last quarter he signed a ₹90,000 "generative engine optimization" retainer. The deliverables were a file called llms.txt, a batch of FAQ schema, and forty pages rewritten into bullet lists. Six weeks later his traffic was flat and his sales team was still cold-calling. He wanted to know how to rank in Google AI Mode. What he bought was a package built on guesswork.
In May 2026, Google published an official AI optimization guide in its Search documentation. It answers the question directly, and the answer annoys a lot of people selling packages like Sameer's.
- Google's own guidance states that optimizing for AI features is not a separate discipline from SEO.
- llms.txt, AI-specific content rewriting and special AI schema are not required.
- Crawlability, useful content and clean technical hygiene remain the levers.
- If an agency sells you an "AI Mode ranking" deliverable that is not one of those, ask what document it comes from.
- The honest answer to how to rank in Google AI Mode is: do ordinary SEO properly, then measure differently.
- What Google Actually Published
- How to Rank in Google AI Mode: The Short Answer
- Four Things the Guidance Says You Do Not Need
- What Is Left, and Why It Still Works
- What Sameer Changed at the Pump Company
- Measuring AI Mode Visibility Without Fooling Yourself
- The Honest Trade-Off
- Questions to Ask Before You Sign an AI Search Retainer
What Google Actually Published
The document lives in Google's Search Central fundamentals section, at developers.google.com/search/docs/fundamentals/ai-optimization-guide. It was published in May 2026. You can read the whole thing in about ten minutes, which is roughly ten minutes more than most people who quote it have spent.
The core position: AI features in Search are surfaced from the same index, using the same signals, as the rest of Search. Google states that AEO and GEO are not disciplines separate from SEO. There is no second door. There is no parallel ranking system you get into with a special file.
How to Rank in Google AI Mode: The Short Answer
Be crawlable. Be genuinely useful on a specific question. Be technically clean enough that a crawler and a person both get the content quickly. That is the whole answer, and it is boring on purpose.
That is how to rank in Google AI Mode, stated in full. Boring is a feature here. Every previous "the rules changed completely" moment in search ended with the same conclusion, and this one has the unusual advantage of being written down by the search engine itself. When a client asks us how to rank in Google AI Mode, we start by auditing the things Google's guidance names, in the order it names them.
Four Things the Guidance Says You Do Not Need
| Sold as | Required? | What to do instead |
|---|---|---|
| An llms.txt file at your domain root | No | Check robots.txt does not block the pages you want read |
| Rewriting pages into "AI-friendly" bullet blocks | No | Write for the person with the problem; structure follows |
| Special AI-only structured data types | No | Use the standard schema that matches your content |
| A separate "GEO strategy" document | No | One search strategy, measured across more surfaces |
| Crawlable, fast, useful pages | Yes | This is the work |
Notice what is missing from the left column: anything a person searching your category would benefit from. That is the tell. None of that means llms.txt is harmful. It means it is optional, unproven for Google Search, and should be priced as an experiment rather than sold as a ranking factor.
How to Rank in Google AI Mode: What Is Left, and Why It Works
What Sameer Changed at the Pump Company
We threw out the deliverables list and rebuilt the plan around the guidance. Every item below maps to something the document names, which is the only defensible way to answer how to rank in Google AI Mode for a real business. Illustrative, but this is the shape of the work.
His product pages listed head, flow rate and motor rating inside a scanned datasheet image. We moved those specifications into a plain HTML table on each page. His site had one enormous "Products" page; we split it into fourteen pages, each answering a buying question a plant engineer would actually type. His robots.txt was blocking a whole directory of application notes, left over from a staging setup nobody had revisited in two years.
No llms.txt. No rewriting into robot-speak. Anyone asking how to rank in Google AI Mode for a niche industrial product is usually one blocked directory and one image-only spec sheet away from a much better answer.
Measuring AI Mode Visibility Without Fooling Yourself
This is where the real change sits. Not in tactics. In reporting. Most teams asking how to rank in Google AI Mode are really asking why their dashboard stopped making sense.
- Track branded search volume over time. Answer surfaces create demand you never see as a click.
- Watch impressions-to-clicks by query group, not sitewide. Informational queries will drift; commercial ones usually hold.
- Add a "how did you hear about us" field to your enquiry form and read the free text weekly.
- Log the questions your sales team is asked repeatedly, then check whether a page of yours answers each one.
- Keep server-log access. Crawler behaviour tells you more than any third-party visibility score.
The Honest Trade-Off
Here is the part agencies skip. Following the guidance well can still cost you clicks. If your page answers a question so cleanly that the answer surface can summarise it, some readers will never arrive. That is a genuine loss, and pretending otherwise is dishonest.
What you gain is different: you get quoted, your brand gets repeated, and the visitors who do click arrive further down the decision. Sameer's enquiry count moved before his session count did. Fewer visitors, better conversations. For a manufacturer selling ₹4 lakh pumps, that trade is fine. For a display-ad-funded content site, it may not be, and no amount of schema fixes that.
Questions to Ask Before You Sign an AI Search Retainer
Sameer's ₹90,000 mistake was not stupidity. He asked a reasonable question, got a confident answer, and had no way to test it. So here is the test. Five questions, asked in a first call, sort the serious vendors from the rest in under ten minutes.
One: which primary source says this is required? Google published guidance in May 2026 on optimizing for AI features in Search. Any proposal for how to rank in Google AI Mode should be traceable to that document, to another search engine's own documentation, or to a test the vendor ran and can describe. Blog posts citing blog posts do not count.
Two: what will you remove? Good search work deletes things. Duplicate tag containers, thin city pages, an old blog category producing nothing. A proposal that is entirely additive is a proposal designed to be billable rather than effective.
Three: how will we know it worked? Ask for the specific report, the specific date, and what a failure would look like. Anyone answering how to rank in Google AI Mode should also be willing to say what they would consider a bad outcome in ninety days.
Four: who writes the content? If the answer is vague, the pages will read like everyone else's, and pages that read like everyone else's do not get cited. Your engineers, your service managers and your sales team hold the details nobody else can publish. Someone has to interview them.
Five: what happens to our own site if we stop paying you? The correct answer is that you keep the pages, the schema, the redirects and the documentation. Anything that evaporates when the retainer ends was rented, not built.
Sameer now runs that list on every marketing proposal, not only search ones. His current agency answered four of the five well and admitted uncertainty on the fifth, which is a better sign than five confident answers. Working out how to rank in Google AI Mode is much easier once you have stopped paying for things nobody can explain.
- Google's May 2026 AI optimization guide says AI features are not a separate SEO discipline.
- llms.txt, AI-specific rewriting and special AI schema are explicitly not required.
- Crawlability, one clear question per page, facts in text and accurate standard schema are the work.
- Change your measurement before you change your tactics.
- Anyone who tells you how to rank in Google AI Mode without citing that document is guessing with your money.
Harsh Virani
Digital marketing and web development expert at DL Minds. Passionate about helping businesses grow through innovative technology solutions and strategic digital marketing.