AI Mode has a billion users. How to get your company cited when search answers instead of listing
Google's AI Mode passed a billion monthly users in May, queries in it are three times longer than a classic search, and the answer now arrives before the links. Most advice about being found by AI is old SEO with a new invoice. Here is what actually changes, and eight things to do about it.
Google said three things about AI Mode at I/O on 19 May that should change how any company thinks about being found. It has passed a billion monthly users worldwide. Queries in it have more than doubled every quarter since launch. And the average AI Mode query is triple the length of a traditional search, with the most common first words now "what", "how", "I", "is" and "can".
The same week, Google made Gemini 3.5 Flash the default model behind AI Mode, merged AI Overviews and AI Mode into one continuous experience, and launched a search box that accepts text, images, files, videos and open Chrome tabs. It also announced information agents that run in the background and keep a topic updated for you, so a chunk of future "searches" will be issued by software on a schedule rather than by a person at a keyboard.
Put plainly: for a growing share of the people who might buy from you, the first thing they see is an answer, and your company either appears inside that answer or it does not. That is a different game from ranking on a page of ten links, and most of what is being sold as "generative engine optimisation" is the old game with a new invoice attached.
What actually changed
Three mechanics matter. The rest is detail.
Questions replaced keywords. A query three times longer than before is a sentence, and a sentence has an intent, a constraint and usually a context. "AI invoice processing for a company with three ERPs in India" is a question a page can answer well or badly. "invoice automation software" was a phrase a page could contain. The unit of relevance has moved from matching words to answering the thing that was asked.
Fan-out replaced the single lookup. AI Mode takes one question and breaks it into several sub-queries, runs them at the same time, and assembles an answer from the pages that best serve each one. Google has described this openly. The practical consequence is that you no longer need to be the best page for the whole question. You need to be the clearest page for one of the sub-questions the system generates on the way to the answer. A page that nails "what does an invoice processing pilot cost" can be cited in an answer to a question that never mentioned cost.
Citation replaced the click. Google added direct links, previews and more visible source references to AI answers in May, which helps. It still changes the funnel. A cited company gets named inside the answer whether or not anyone clicks. An uncited company with a good ranking underneath the answer is on page two in everything but name. The measurement you have been using, clicks from search, now under-counts the influence you have and hides the influence you have lost.
Eight things that move the needle
None of these are tricks. They are the habits of a page that is easy to quote accurately.
Lead with the answer. For every question a page is meant to serve, the answer should appear in the first 60 words after the heading, in plain prose, with the specific noun or number in it. The explanation can follow. A model assembling an answer from twelve sources takes the sentence that states the thing, and skips the paragraph that builds up to it.
One page per sub-question, in a cluster. Take your main topic and write down the sub-questions a careful buyer would ask on the way to a decision: what it costs, how long it takes, what can go wrong, how it compares, what happens after. Give each one its own section or page with a heading phrased the way the question is asked. This is how you become the source for a fan-out query you never saw.
Put entities, numbers and dates in the text. A model cannot read a number that lives inside an image or a chart. Prices, timelines, percentages, product names, locations and dates belong in sentences. And they should be the same everywhere: the price on your pricing page, your comparison page and your partner's directory listing should match, because inconsistency reads as unreliability.
Mark it up. Article, FAQ and Organization structured data, with real publication and modification dates. It costs an afternoon of engineering, and it is the difference between a system parsing your page and guessing at it. The "frequently asked" section at the bottom of every article on this site exists for exactly this reason.
Say something that is not on any other page. Fan-out rewards the source that adds a fact the others do not have. Your own numbers, your own process, a failure you saw and what caused it. A page that restates the consensus is one of a hundred candidates. A page with a first-hand detail is the candidate that gets quoted.
Keep it current, and prove it. Background agents check whether a page has changed. A modified date that moves when the content does is a signal; a page last touched in 2024 is a page that will be skipped when a fresher one exists. Set a quarterly review on any page you want cited, and change the substance when you touch it, not only the date.
Get mentioned in places you do not control. Models weigh brand mentions across the web, including unlinked ones, when deciding whether a company is a credible answer. Directories, partner pages, podcast show notes, conference speaker lists and customer case studies all count. Decide on one sentence that describes what your company does and make sure it appears in the same form everywhere, so that the mentions reinforce each other rather than describing five different companies.
Measure citations, not only clicks. Write down the twenty questions a buyer asks on the way to choosing a company like yours. Once a month, run them through AI Mode, ChatGPT, Claude and Perplexity, and log whether you are named, which page is cited, and who is named instead. That log is your ranking report now. It takes an hour and it tells you more than your search console does.
What to stop doing
Stop publishing pages that exist to contain a keyword. They were marginal before and they are invisible now, because they answer nothing.
Hidden pricing is the next casualty. A page that says "contact us" cannot be cited in an answer to "what does this cost", so the answer gets assembled from competitors who published a number.
And the blog cannot stay a separate department from the product. The pages most likely to be cited are the ones that describe, in specific terms, what you actually do and what happens when you do it. Marketing that avoids specifics is marketing that avoids being quoted.
What this means for a company selling to businesses
Enterprise and startup buyers now do their first pass of vendor research in an AI answer. They ask a long, specific question, they read a synthesis, and they take the two or three names inside it to the next stage. If you are in that synthesis you get evaluated. If you are not, the RFP arrives with somebody else's assumptions baked in.
This is one reason our own use-case library is written as answers to questions rather than as feature lists: each page states what the use case is, what it costs, how long it takes and where it fails, in text, with dates. It is the same AI strategy advice we give clients about their own sites, applied to ours.
Frequently asked
How do you get cited in Google AI Mode? By being the clearest source for one of the sub-questions AI Mode generates when it breaks a query apart. In practice that means stating the answer in the first 60 words after a heading, keeping numbers, names and dates in the text rather than in images, marking pages up with Article and FAQ structured data, updating the substance regularly, and adding facts that appear nowhere else.
Is generative engine optimisation different from SEO? The foundations overlap: crawlable pages, clear structure, real authority. The difference is the unit of success. Classic SEO aims to rank a page for a keyword. In AI search you aim to be cited for answering a question, often a sub-question the user never typed, and being named inside the answer matters even when no click follows.
How should a company measure visibility in AI search? Keep a list of the twenty questions a buyer asks on the way to choosing you, run them monthly through AI Mode, ChatGPT, Claude and Perplexity, and record whether you are named, which page is cited and which competitors appear. Track that alongside clicks, because clicks alone now under-count both the influence you have and the influence you have lost.
Related reading
- Gemini Spark works while your laptop is closed
- Enterprise AI transformation: what it actually means
- How to scope an AI deployment in two weeks
A billion people a month are now asking a machine a full sentence and reading the reply. The companies that get named in that reply are the ones whose pages answer a specific question with a specific fact. Write those pages.