Field Guide No. 15 · AI Search
Google AI Mode has a billion users. Here’s how to get cited.
AI Mode rarely bothers with a list of blue links anymore. You ask it something, it quietly splits that into a dozen smaller questions, and it builds one answer out of whatever passages it finds for each. Getting into that answer is its own kind of work, and most sites are still optimizing for the thing it replaced.
At I/O this past May, Google mentioned almost in passing that AI Mode had crossed a billion monthly users. A year before that it barely existed; back in December it was somewhere around 75 million. So the feature a lot of people still mentally file under “experiment” is now where a real slice of your customers go to look things up, and that slice keeps growing.
What throws people is how it decides who to quote. I’ve watched a page that sits comfortably on page one of normal Google get left out of the AI Mode answer completely, while something stranded back on page three gets pulled in. Once you see why that happens, everything else here follows.
If you read my piece on getting cited by Perplexity, some of this will sound familiar. You’re trying to be a source the model reaches for, not to win a position. AI Mode is the same bet, just playing out across Google’s traffic instead of Perplexity’s smaller pool.
The mechanic
How a single query becomes a dozen
The mechanism has a name: query fan-out. Ask AI Mode one thing and it spins up a handful of related searches behind the scenes, runs them together, and stitches a single answer out of the bits it gets back. It’s easier to follow with a real question in front of you.
Query Fan-Out
One question in, many searches run
Google quietly fans it out into
Which is why your exact ranking matters less than it once did. When Ahrefs looked at Google’s AI citations, only about 38% came from pages in the top 10, so the clear majority were quoted from further down (and that top-10 share has been falling). The reason is that it’s grabbing passages, not handing out trophies for position. Answer a few of those sub-questions well and you can get pulled into the same response from several different angles. Put one keyword on one thin page and you’ve got a single shot, which usually misses. It’s the same reason topic clusters have quietly overtaken single keywords.
Forget ranking ten links and picking a winner. AI Mode carves your question up, answers each piece from whatever it can find, and credits those pages. The job is to be the best answer to one of the small questions, not the best page for the big keyword.
What it rewards
The signals that actually move citations
Google’s own line on this is almost annoyingly plain: optimizing for AI Mode is just SEO. No hidden setting, nothing to toggle. What does shift is where the weight falls, and from what I’ve seen getting clients quoted, it falls roughly like this.
Relative Emphasis
What gets a passage pulled
Extractable answers
A direct answer in the first sentence of a section. Definition-first, question-then-answer structure pulls best during fan-out.
Sub-intent coverage
Pages that answer several related questions create more chances to be the quoted passage across different branches.
Freshness
Across AI assistants, cited pages run about 25.7% fresher than organic results (Ahrefs). A visible, recent date and genuine refreshes keep you in range.
First-party authority
Recent core updates tilted toward brand-owned and official sources. Being the primary source matters more, not less.
Real experience
Demonstrated, first-hand experience beats manufactured expertise. Google favors content from people who did the thing.
Technical fundamentals
Crawlable, fast, parseable. If retrieval can’t reach your passages, nothing above matters. Technical SEO for AI.
How to rank in AI Mode
Five moves, in order of impact
Map and answer the sub-questions
For each important page, list the 8 to 12 questions a real buyer asks around the topic, then make sure each is answered cleanly on the page or across a tight cluster. That coverage is exactly what fan-out rewards.
Lead every section with the answer
Put the answer in the first sentence under each heading, and make headings match the questions people actually ask. Make that opening line quotable on its own.
Refresh content and show the date
Add visible dates, then genuinely refresh your top pages with new data and current claims. Fresher pages get cited disproportionately. Here’s the freshness framework I use.
Build first-party authority
Original data, first-hand experience, and mentions in real publications all lift the authority signal the core updates and AI layer reward. See building topical authority.
Measure where you already show up
Before changing everything, see how you and competitors actually appear across AI engines. Start with a competitor AI visibility analysis.
Spend wisely
Worth your time vs. busywork
Worth your time
- ✓ Rewriting openers to answer the question first
- ✓ Covering the full set of sub-questions
- ✓ Real content refreshes with visible dates
- ✓ Standard, accurate structured data
- ✓ Earning genuine third-party mentions
Skip it
- ✕ llms.txt files (Google signals it doesn’t use them)
- ✕ “Chunking” content into odd fragments
- ✕ New schema invented just for AI
- ✕ Keyword-stuffed pages built only to rank
- ✕ Chasing every GEO hack of the week
Side by side
AI Mode vs the other answer engines
Please don’t build three separate strategies for this. AI Mode, AI Overviews, and Perplexity reward more or less the same things; they just turn the dials to different settings. If AI Overviews are where you care most, my GEO playbook goes deeper on that one.
| Signal | AI Mode | AI Overviews | Perplexity |
|---|---|---|---|
| Query fan-out depth | High | Medium | Medium |
| Sub-intent reward | High | Medium | High |
| Freshness weight | High | Medium | High |
| Domain authority | Medium | High | Medium |
| Reach / scale | High | High | Lower |
Levels are relative estimates from observed citation patterns, not published platform data.
Where AI Mode pulls ahead is sheer reach and how hard it fans out. A billion users, most of the citations coming from outside the top 10: that leaves room for sites that genuinely cover their topic, not just the ones parked at #1 for the headline term. It’s the same point I got into in the zero-click piece.
The bigger pattern
The recent core updates point the same way
None of this is happening in a vacuum either. The May 2026 core update wrapped up at the start of June, right on the heels of a turbulent March update where, by Search Engine Land’s analysis, nearly 80% of top-three results shifted positions. In verticals like health, the pattern was clear: clinical and specialist sources gained while broad consumer aggregators slipped.
It all points the same way. Google would rather cite where the information came from than the site repackaging it. If you’re the actual authority on your subject, the core algorithm and the AI layer are both drifting toward you. If the plan was to out-publish everyone with thin posts, well, they’re drifting the other way.
Questions
Frequently asked
What is Google AI Mode?
Does ranking #1 still matter for AI Mode?
How is AI Mode different from AI Overviews?
Do I need new schema or an llms.txt file?
What is query fan-out?
How do I rank in Google AI Mode?
Sources
- AI Mode usage (1B+ users, ~75M in December 2025, May 2025 launch), Google I/O 2026, reported by Search Engine Journal.
- AI Overviews at ~2 billion monthly users, Google, reported by Digiday.
- AI citation distribution (~38% from top-10 pages), Ahrefs.
- AI citations ~25.7% fresher than organic results, Ahrefs.
- March and May 2026 core update volatility, Search Engine Land.
Want to know where you stand in AI search?
Send me your site and I’ll show you where you’re getting cited, where you’re invisible, and the fastest fixes. Free review, one business day, no sales pitch.
Get a free AI visibility review →