If you’ve been on LinkedIn in the past year you’ve definitely seen something along the lines of: “Your competitors are starting to show up in Google’s AI Overviews and you’re not. You’ll definitely need an AI SEO retainer and an llm.txt file”

Since AI search features started rolling out, the whole industry has been essentially best-guessing how they pick sources, Distl included. It was all we had until 15 May 2026, Google Search Central published the first official guide to optimising for AI Overviews and AI Mode. Several of those common tactics are now off the table.

What Google Confirmed Last Week

AI Overviews are powered by two mechanisms:

Retrieval-augmented generation is the first. Google calls it grounding. In plain English: a question triggers an AI Overview, the system runs a normal Google search, and picks the most relevant pages from the index. It then rewrites the useful bits into an answer with clickable citations back to the source pages. AI Overviews are a presentation layer over the same search results you already compete in.

Query fan-out comes next. When a question is complex, the system runs several related searches in parallel and stitches the results together. Someone asking “how do I fix a lawn full of weeds” might trigger fan-out queries like “best herbicides for lawns” or “remove weeds without chemicals”. These run concurrently in the background.

Both mechanisms sit on top of Google’s existing ranking system. Google writes:

“From Google Search’s perspective, optimising for generative AI search is optimising for the search experience, and thus still SEO.”

The system that picks the sources for an AI Overview is the same one that picks the top ten blue links. If your page can’t rank, it can’t be retrieved, and it can’t be cited.

A visual guide of what Google confirmed is required, helpful and optional for optimising towards AI search.

Six Common Tactics the Guide Ruled Out

The guide includes a section called “Mythbusting generative AI search: what you don’t need to do”. Six items on that list have shown up in AEO advice over the past 12 to 18 months.

llms.txt files. Probably the most-discussed AEO tactic of the past year. Google: “you don’t need to create new machine readable files, AI text files, markup, or Markdown to appear in generative AI search.”

Chunking content for AI. The idea was that you should break long articles into bite-sized pieces because that’s how LLMs read them. Google: “Google systems are able to understand the nuance of multiple topics on a page and show the relevant piece to users.” You’re likely better off being comprehensive and clear with your content, rather than chunky. The retrieval model reads pages the same way as the rest of Search.

Rewriting content for AI. Google: “AI systems can understand synonyms and general meanings of what someone is seeking, in order to connect them with content that might not use the same precise words.” Write one set of pages well for the humans reading them.

Citation farming and inauthentic mentions. Google: “seeking inauthentic mentions across the web isn’t as helpful as it might seem.” Authority earned through useful content beats authority manufactured at scale.

Over-investing in structured data for AI. Schema is good for normal SEO rich results. For AI Overviews: “Structured data isn’t required for generative AI search, and there’s no AI specific schema.org markup you need to add.”

Publishing fan-out variation pages. Targeting every possible fan-out query with its own page. Doing this “primarily to manipulate rankings or generative AI responses in Google Search violates Google’s scaled content abuse spam policy.” This is the only thing that could be actively hurting your results, many of Google’s core algorithm updates are based around penalising spam and pages made for rankings and not for actual useful content.

None of these tactics were necessarily scams. They were reasonable guesses when the mechanism behind AI Overviews was opaque. With the mechanism now published, the working list of what helps has gotten clearer. A lot of the effort going into the items above can move to the items below.

Commodity vs Non-Commodity Content

A useful new phrase in the guide is “non-commodity content”. Google contrasts two example articles to make the point. One is titled “7 Tips for First-Time Homebuyers”. The other is “Why We Waived the Inspection & Saved Money: A Look Inside the Sewer Line”.

Commodity content is the first kind, anyone could write it and forty other sites already have. A retrieval model can synthesise an answer from the existing pages on its own, without ever needing to cite the new one.

Non-commodity content can’t be synthesised. The key word is ‘We’, it’s a unique take or action that is specific to the writer. It’s a specific person describing a specific decision with specific consequences. The model needs the page because the model can’t make up the story.

Generic listicles are the new thin content. The pages that get cited in AI Overviews are the ones that contain information the model couldn’t generate without them.

Instead of briefing topics, brief experiences: ask the team what they’ve seen, fixed, lost money on, or argued about with clients. Their working knowledge is invaluable.

What the Guide Recommends Instead

The list Google does publish, ranked by impact, looks a lot like good SEO with two surfaces most AEO conversations don’t cover.

First, content that comes from someone who’s done the work. If you can’t write something that wouldn’t exist without you, the page isn’t worth publishing. Everything else on this list is downstream of that one.
Technical SEO foundations come next. To appear in an AI Overview, the page has to be eligible to appear in Google Search at all. Indexable, crawlable, fast, with semantic HTML that screen readers and browser agents can parse. If the basics aren’t right, nothing else matters.
A Google Business Profile is AI surface optimisation for local businesses. A Merchant Center feed is the same for ecommerce. Google explicitly says AI responses can include both. The same setup work that’s always served local SEO and Shopping is now also serving generative AI search.
High-quality images and video where they support the content are still worth the effort. AI Overviews bring in media alongside links, and the image and video SEO that’s worked for years continues to work.
Last on the list, for businesses with budget and curiosity to spare: agentic experiences. Browser agents read your DOM and accessibility tree to perform tasks. New protocols like Universal Commerce Protocol let search agents transact with your store. For most Australian SMEs, none of this is a line item for 2026.

What Google Didn’t Change

A common narrative over the past year has been that AI changed everything and SEO needs to be rebuilt from scratch. Ranking systems work the same way as before, and the retrieval layer that scores them improved. The next 12 months will reward businesses that double down on the basics while the rest of the field debates new acronyms.

If you’d like a take on what your site needs (and just as importantly, what it doesn’t), talk to the Distl SEO agency in Perth.

Sources: Google’s AI Optimisation guide (last updated 15 May 2026)

Google’s Guidance on Using Generative AI Content (last updated 10 December 2025).