Google Says Don't Format for AI. The Math Disagrees.
Danny Sullivan, Google's Search Liaison, went on a podcast and told you to stop cutting your content into chunks for AI. The same week, every system that decides whether your page shows up inside an AI answer kept reading your content one passage at a time, the way it always has. Both things are true. That contradiction is costing you citations right now, and this article proves the resolution is not "write naturally" or "format for robots," it is clear structure that serves a skimming human and a passage-hungry machine at the same time, with no rebuild required.
The click is drying up, so getting quoted is the whole game now
The ground moved while owners were arguing about tone. In the first four months of 2026, 68.01% of US Google searches ended without anyone clicking through to a website, according to Rand Fishkin at SparkToro using Similarweb data, published June 8 2026. That number was 60.45% in 2024. Round it however you like. Roughly two out of three times someone searches for what you sell, they read the answer on Google's page and leave.
It gets sharper when an AI summary sits at the top. Ahrefs analyzed 300,000 keywords and found that AI Overviews cut the clickthrough rate of the top-ranking page by about 58% in December 2025, up from a 34.5% reduction in April 2025, per Ryan Law's February 4 2026 report. If you were the number-one result pulling 1,000 visits a month from a query, an AI Overview on that query now quietly takes roughly 580 of them. The pain scales down the page: about -50% at position two, -46% at position three, sliding to around -19% at position ten. Lower rank, smaller cut, but a cut everywhere.
Pew Research Center measured what people actually do when a summary appears. Looking at the browsing data of US adults in March 2025, published July 22 2025, they found users clicked a traditional organic result in just 8% of visits when an AI summary was present, versus 15% when it was not, and clicked a link inside the summary only about 1% of the time. Half your already-shrunken click rate, gone the moment the box appears. The visit you used to win by ranking first is now a visit someone spends reading a 67-word summary, the median length Pew measured, that cites three or more sources 88% of the time.
If you are one of those three cited sources, you exist in the answer. If you are not, you are invisible to two-thirds of your market. The fight stopped being about clicks alone. It became about being quoted.
Owners heard "just write naturally" and stopped doing the one thing that earns citations
A roofer in Tampa reads that Google's own liaison says don't format for AI, exhales, and tells his marketing person to stop fussing over headings and "just write good content." Reasonable on its face. He trusts the source. Google said it, so it must protect him. What he actually did was remove the structure that lets a machine lift a clean answer about flat-roof repair costs out of his page and put his name next to it.
The confusion is real because the advice arrived with authority. The quote that launched a thousand bad decisions comes from the Search Off the Record podcast, quoted by Mike King at iPullRank on January 15 2026:
"One of the things I keep seeing over and over... is that turn your content into bite-sized chunks, because LLMs like things that are really bite size, right? So we don't want you to do that. I was talking to some engineers about that. We don't want you to do that. We really don't. We don't want people to have to be crafting anything for Search specifically."
Danny Sullivan, Google Search Liaison, Search Off the Record podcast, via iPullRank, January 15 2026.
Read it twice and you notice what he is actually objecting to. Crafting content "for Search specifically." Gaming. Writing for the machine instead of the person. He is not saying structure is bad. He is saying do not turn your page into robot food. Owners heard the second half and deleted the first. They collapsed "don't write artificially for AI" into "don't organize your thinking," and those are not the same instruction. One protects your reader. The other abandons them.
This is the third time I have watched a "stop optimizing, just write" message get weaponized into an excuse to publish unstructured mush. In seventeen years the pattern never changes. A platform says something sensible about not gaming the system, and a swath of businesses hear permission to stop thinking about how anyone, human or machine, will find the answer inside their work.
The systems read passages, which is why the math overrules the advice
Sullivan's intent and the machine's mechanics are two different layers, and only one of them decides whether you get cited. Mike King put the resolution plainly in that same January 15 2026 iPullRank piece:
"Chunking and creating content for users are not mutually exclusive."
Mike King, iPullRank, January 15 2026.
He goes further, and this is the load-bearing line for everything that follows:
"In every one of these systems, passages remain the atomic unit of meaning."
Mike King, iPullRank, January 15 2026.
Atomic unit means the smallest piece the system works with. Retrieval systems behind AI answers do not grade your whole page as one block. They break it into passages, score each one against the question, and pull the passage that answers best. A page that buries its answer to "how much does flat-roof repair cost in Tampa" inside paragraph nine of an unbroken essay is a page where the strongest passage is hard to isolate. Clear sections with self-contained answers are easier to lift. That is not crafting for Search specifically. That is writing a paragraph that stands on its own, which a human skimming on a phone wanted anyway.
The two audiences converge instead of competing. A business owner scanning your page between meetings and a retrieval system scoring passages both reward the same thing: a section whose meaning survives being read alone. I have written before that getting cited is precise SEO, not a new discipline, and this is the proof. You are not learning a new skill. You are doing the old skill with more discipline. The formats that actually earn quotes are specific, and worth knowing if you want to understand which content AI engines reach for and which they ignore.
There is a second reason structure alone will not save you, and it keeps this honest. Rand Fishkin, SparkToro, June 8 2026:
"Much as it pains me to say, there's not much point (nor any hope) of fighting back by simply getting better at SEO. Don't take that the wrong way, your SEO still matters as much or more than ever before, it just won't earn you traffic the way it once did."
Rand Fishkin, SparkToro, June 8 2026.
So structure earns the citation. The citation earns visibility, brand recall, and the occasional click, not the traffic flood of 2019. Cyrus Shepard's Zyppy study of 400-plus sites, published April 9 2026 and covered by Search Engine Journal April 10 2026, found the sites gaining Google traffic in 2026 shared five traits: they sell a product or service, let users complete a task, own proprietary assets, focus tightly on one topic, and have a real brand. Sites with all five won traffic about 69.7% of the time. Sites with none won 13.5% of the time. Ninety-two percent of the winners owned proprietary assets, against 57% of the losers. Structure makes your proprietary assets liftable. The assets themselves are what make you worth lifting.
Five moves you can do, delegate, or hand your agency as exact questions
None of this is a rebuild. It is editing discipline applied to pages you already have, starting with the ten that matter most to revenue.
First, give every section a heading that states a claim or answers a question. Not "Pricing" but "Flat-roof repair in Tampa runs $400 to $1,200." A reader skimming your headings should get your whole argument, and a retrieval system reads those headings as signposts to where each answer lives. Do this yourself in twenty minutes per page. If you delegate it, the exact instruction is: "Rewrite every H2 and H3 on these ten pages so each one makes a specific claim instead of naming a topic."
Second, make each section answer its own question in the first two sentences, then expand. The buried answer is the citation you lose. Open the section with the payoff, then give the context, the caveats, the story. A human gets what they came for fast. A machine finds a clean, self-contained passage to quote. Ask your writer or agency this directly: "Does the first sentence of each section answer the heading's implied question without needing the rest of the page for context?"
Third, write your factual claims as complete, standalone sentences. "It costs more in winter" means nothing lifted out of context. "Flat-roof repair costs 15 to 20 percent more in winter because cold makes membrane adhesives harder to apply" survives on its own in an AI answer with your name attached. Every number, every price, every specific gets the full sentence around it. This is the single highest-impact edit, and it costs nothing but attention.
Fourth, put your proprietary asset on the page in liftable form. Shepard's data says 92% of traffic winners own proprietary assets. Your real pricing, your job-site photos with specifics, your warranty terms, your actual case numbers, the thing only you have because you do the work. State it plainly in its own section. A generic page restating what ten competitors already said gives a machine no reason to pick you. The question for your team: "What can we say on this page that no competitor can copy, and is it written as a clean, quotable section?" If you want the longer view on how engines filter before they ever cite you, the stages AI search runs content through before quoting it are worth walking once.
Fifth, stop crafting anything that reads like robot food, which is the part Sullivan was right about. No keyword-stuffed FAQ blocks written for a parser, no fake question headings nobody would type, no padding to hit a word count. Clear structure and natural language are the same edit done well. If a sentence would embarrass you read aloud to the roofer's customer, it will not help you with the machine either. The two standards are the same standard.
Measure citations and answer presence, not the traffic line you keep staring at
The vanity metric that will mislead you for the next year is your organic traffic graph. It is going down, and structure will not reverse it, because the click itself is what is disappearing. Fishkin's data is the calm-down: AI tools send less than 1% of all referral traffic out, and Google's AI Mode accounted for only 0.34% of searches January through April 2026. The traffic is not flooding to AI referrals. It is evaporating into zero-click answers. So judging your content by the traffic line is grading the wrong test, which is exactly why I keep telling people to stop scoring content by Google traffic alone.
Measure three things instead. One, citation presence: search your top ten money queries in AI Overviews and an AI assistant once a month, and write down whether you are named among the cited sources. That binary, cited or not, is the new ranking. Two, answer accuracy: when you are cited, is the summary repeating your specific number, your price, your claim, or a competitor's. Being cited with someone else's facts attached is a problem structure can fix. Three, assisted conversions: are people arriving already knowing your price, your terms, your name, because they read it in an answer before they ever typed your URL. That is the citation paying off downstream, invisible to your traffic report.
Give it ninety days before you judge any of it. Citations move slower than rankings did, and the systems recrawl and re-evaluate on their own schedule. Check monthly, decide quarterly. One more trap: do not celebrate a citation that quotes a vague, generic line from your page. That citation does nothing for your brand because it could have come from anyone. The citation worth winning quotes the specific, proprietary thing only you could say.
Structure is one part of a much bigger job, and I cover where it fits in my guide to AI search visibility.
Frequently Asked Questions
Did Google actually tell people not to format content for AI?
Yes, sort of, and the wording matters. On the Search Off the Record podcast, quoted by Mike King at iPullRank on January 15 2026, Danny Sullivan said Google does not want people turning content into bite-sized chunks or "crafting anything for Search specifically." Read closely, he is objecting to writing artificially for the machine, not to organizing your content with clear sections and headings. Those are two different things, and owners keep collapsing them into one.
If I structure my content clearly, will my Google traffic come back?
No, and anyone promising that is selling you something. Rand Fishkin's SparkToro data from June 8 2026 shows 68.01% of US searches now end without a click, and AI tools are not sending that traffic anywhere else yet. Clear structure earns you citations inside AI answers, which protects your brand visibility and feeds assisted conversions. It does not rebuild the click volume of five years ago, because the click itself is disappearing from the search results page.
What is the single most important change to make first?
Rewrite your factual claims as complete, standalone sentences on your ten highest-revenue pages. A sentence like "it costs more in winter" cannot be lifted into an AI answer with your name attached, but "flat-roof repair costs 15 to 20 percent more in winter because cold makes adhesives harder to apply" can. Every price, number, and specific claim should survive being read alone. This costs nothing but attention and gives a retrieval system a clean passage to quote.
You spent years learning that ranking first was the prize. It still helps, but the top-ranking page now loses 58% of its clicks to an Overview, per Ahrefs in February 2026, so first place is no longer the finish line you trained for. The new prize is being one of the three or more sources a 67-word summary quotes, and that comes down to whether your best answer can stand on its own when a machine reaches in and pulls it out. Structure decides that. Which means the question is no longer whether you rank for the query, but whether your page even survives being read one paragraph at a time, the way every machine has been reading it all along.
