You wrote a good page.
Not a thin, keyword-stuffed page. A real one. You explained the thing you actually know, in your own words, because you were tired of watching customers get bad advice about it. You were pleased with it.
Then a customer asked an AI tool the exact question that page answers. And the answer came back quoting somebody else.
That stings in a particular way, because you cannot even argue with the quality. Your page is good. It just was not the one that got picked. And you have no idea why.
Here is why. You built that page for people. The models that now sit between you and your customer read a page very differently, and almost nothing you were told about SEO prepares you for it. The good news is that the fix is structural, not creative. You do not need to write more. You need to make one part of the page easy to lift.
Why does AI quote a competitor instead of your page?
Because your page gives it nothing clean to pick up. AI tools do not read your page top to bottom and form an opinion. They go looking for a specific passage that answers the question, and they quote that. If no passage on your page is easy to lift, you get skipped, even when your page is the better page.
Sit with that for a second, because it inverts most of what you have been told.
Traditional SEO was a competition between whole pages. Google looked at your page as one object, scored it, and ranked it against other objects. So the advice made sense: cover the topic thoroughly, build authority, be the most complete resource on the subject.
An AI answer engine is not running that competition. It is assembling a short answer for one person who asked one thing. It needs a couple of sentences it can quote with confidence. Your 2,000 word masterpiece is not a candidate. A passage inside it is.
And most business pages are built so that no such passage exists. Watch how a typical service page opens:
- A headline that names the business, not the question.
- A warm paragraph about how long you have been doing this.
- A paragraph about your approach and your values.
- Then, about 600 words in, buried in the middle of a section, the actual answer. A human reader skims past the first three and finds the fourth. They are patient with you, because they arrived on purpose. A model has no patience and no purpose beyond the question in front of it. It looks for the answer, does not find one sitting in plain view, and moves to a page where the answer is right there.
Nothing is broken. Nothing got penalised. You simply built a page for a reader who scrolls, at a moment when the thing choosing between you does not scroll at all.
The real cost: fewer clicks, handed to whoever got quoted
The reason this matters more than it did two years ago is that the clicks are shrinking, and the ones left over go to the source that got quoted.
The Pew Research Center tracked what people actually do when an AI summary appears in their search results. This was not a survey of what people think they do. Pew followed the real browsing behaviour of 900 US adults across 68,879 searches, and published the results in July 2025.
The numbers are blunt:
| What the searcher saw | Clicked through to a result | | --- | --- | | An AI summary at the top | 8% of visits | | No AI summary | 15% of visits |
So when an AI summary shows up, the chance of anyone clicking a normal search result roughly halves. And Pew found something else worth knowing: only about 1% of visits ended with someone clicking a link inside the summary itself. The summary is not a polite shopfront that sends people on to you. For most searchers, it is the destination.
This is not a rare event any more. In March 2025, Pew found that about 18% of searches returned an AI summary, so roughly one search in five. Semrush, looking at a much larger keyword set in the same month, found that 13.14% of US desktop queries triggered an AI Overview.
Put the two findings together and you get the situation small businesses are actually in. A meaningful slice of searches now ends with a summary. When it does, far fewer people click anything. And the attention that does survive goes to whoever the model decided to quote.
That is the cost of being unquotable. It is not a ranking penalty. It is quieter than that. You keep your position, you keep your traffic report, and you slowly stop being part of the conversation at the moment a customer is deciding what to buy.
The businesses winning here are not always the biggest ones in the category. They are the ones whose pages are easy to lift from.
How does AI actually read your page?
It reads in passages, not pages. AI search tools break your content into chunks and retrieve the chunk that best matches the question, so the unit that competes is a passage, not the whole article. Discovered Labs, writing on how retrieval works for AI search in 2025, found that leading with a plain two to three sentence answer near the top makes a page significantly easier to retrieve and cite.
That single finding is the whole game, so it is worth translating into plain terms.
Imagine your page has been cut into index cards. Each card holds a paragraph or two. When someone asks a question, the model sorts through everyone's cards, yours and your competitors', and picks the handful that answer it best. Then it writes an answer from those cards and credits where they came from.
Three things follow from that, and they change how you write:
- Your best card competes on its own. It does not get carried by the rest of the page. A brilliant article with a vague opening can lose to a plainer page that answers the question in the first paragraph.
- A card that hedges gets skipped. "It depends on a range of factors" is not an answer. It is a card the model cannot use.
- The card near the top usually wins. Not because of some hidden rule about position, but because that is where writers put the summary, and a summary is the most liftable thing on any page. Most business pages have no card worth picking. They have context, warmth, background, and then eventually an answer that only makes sense if you read the three paragraphs before it. That is good writing for a magazine. It is invisible writing for an answer engine.
So the fix is not to write differently for the whole page. It is to add one card that is obviously, deliberately liftable, and then give the model reasons to trust it.
The anatomy of a page AI actually quotes
Here is what a citable page has that an ordinary good page does not. There are only two parts to it, and neither one takes a developer.
Start with a "Key takeaways" box
Put a short "Key takeaways" box right at the top of the page, above the long explanation. Two or three plain sentences that answer the question the page exists to answer. That box is the piece a model lifts.
This is the pattern Foundation identified when they pulled apart the pages that AI tools cite most often in the data catalog category. The top-cited guides did not open with brand storytelling. They opened with a boxed summary that answered the question outright, in ordinary language, before any of the detail arrived.
Three rules make the difference between a takeaways box that works and one that is just decoration:
- Answer the question, do not preview the article. "In this guide we will look at the main considerations around pricing" is a table of contents. "Most small business websites in Australia cost between X and Y, and the price is driven mainly by how many pages you need and whether the content is written for you" is an answer. Only one of those can be quoted.
- Write it so it survives on its own. Assume the reader will never see the paragraph before or after it. No "as we mentioned above", no "this", no pronouns pointing at something off the card.
- Keep it plain. Short sentences, no jargon, no clever phrasing. You are writing the sentence you would say out loud if a customer rang and asked. That sentence is easy for a person to understand and easy for a model to quote, for exactly the same reason. If writing that plainly feels harder than writing the article did, you are not doing it wrong. Compressing what you know into three honest sentences is the hardest paragraph on the page. It is also the only one with a real chance of being quoted, so it deserves the time. Our guide to writing copy that gets results covers the same discipline for the rest of the page.
Add the trust signals
A liftable answer is only half of it. The model also has to decide whether your answer is worth repeating. That is what the trust signals do, and they are startlingly ordinary:
- A real byline with a real name. A named author, ideally with a line about why that person knows the subject. "By the Team" tells a reader nothing and tells a model nothing.
- A published date and a last updated date. Both, visible on the page. On a subject that moves, an undated page is a page nobody can vouch for.
- Links to the sources you used. If you cite a number, link the study it came from. If you are quoting an industry body, link them. Foundation puts the logic well: each element gives a reader a reason to trust the page before scrolling, and gives a model a reason to quote it.
That framing is the useful part. These are not tricks aimed at machines. They are the same signals a careful human uses when they are deciding whether to believe a page they have never seen before. Who wrote this? When? What is it based on? You already apply that test dozens of times a week without noticing.
One caution, because this area attracts a lot of confident nonsense. Google's E-E-A-T guidance describes what its quality raters look for. It is not a ranking factor you can switch on, and anyone selling you an "E-E-A-T service" is selling you a story. A named byline and honest dates are worth doing because they make the page more trustworthy and more quotable, not because they move a dial in an algorithm.
Why this beats chasing rank
Structure can win on its own. Foundation found an open-source guide sitting at position 35 in search results, which is deep into page four, where essentially nobody goes. That page still earned nine AI citations. It was not winning the ranking competition. It was winning the quoting competition, because it was built to be quoted.
For a small business, that is the most encouraging number in this entire article.
Climbing from position 35 to position 3 is a project. It takes months, money, and links you may never get, and you are competing with businesses who have full-time teams pointed at exactly that. Making your page quotable is an afternoon. It is a box, a byline, two dates and a few links.
The two are not in conflict, and none of this replaces the basics. Your page still has to load properly, be findable, and say something true and useful, which is a whole separate discipline covered in our guide to building a website. But if you have already done the hard work of writing something genuinely good, the structural fix is the cheapest visibility you will buy this year.
One honest caveat, because you deserve it. A takeaways box does not guarantee a citation. Nobody has isolated how much a single box lifts your odds, and any figure you see claiming to is invented. What the evidence supports is narrower and still worth acting on: pages built this way are easier to retrieve and easier to quote, and pages that are not built this way are frequently skipped no matter how good they are.
Do this today: add a takeaways box to your most important page
One page. Ten minutes. Do it before this gets filed under "someday".
- Pick your most important page. Not your homepage. The page that answers the question customers actually ask before they buy: what it costs, how it works, which option suits them, how long it takes.
- Write down the question that page exists to answer. In the customer's words, not yours. If you cannot write the question in one line, the page is trying to do too many jobs and that is your real problem.
- Answer it in two or three sentences. Plain language. Real specifics. No hedging, no "it depends" without saying what it depends on.
- Put it at the top, under a heading that says "Key takeaways". Above the story, above the intro, above everything. Yes, above the bit you are proud of.
- Add the trust signals. A named byline. The published date and the last updated date. Links to anything you cited.
- Read the box on its own. Cover the rest of the page. Does it still make sense, and does it still answer the question? If it needs the paragraph above it, rewrite it so it does not. That is the whole job. You have just given the models something to quote.
Then do the unglamorous part: repeat it on your next four most important pages, and put the last updated date on your calendar so it stays honest. A stale date is worse than no date. Then check your work. Ask the assistants your main question again in a month and see whether the box you wrote is what comes back. That checking loop is what our AI visibility work does at scale.
It is also worth making sure the traffic that does arrive has somewhere to go. Being quoted brings you a smaller number of much warmer visitors, people who have already been told you are the credible source. If the page they land on does not give them an obvious next step, you have won the citation and lost the customer. Our guide to conversion rate optimisation walks through how to find the weak points in that path.
Questions people ask about getting cited by AI
How do I get my content cited by AI tools?
Give the model a passage it can lift, then give it reasons to trust that passage. In practice: a "Key takeaways" box at the top with two or three plain sentences answering the page's main question, plus a named author byline, a visible published and last updated date, and links to the sources you cited. AI tools retrieve passages rather than whole pages, so the passage is what competes.
Do I need to rank number one to be quoted by AI?
No. Ranking helps, but structure can carry a page on its own. Foundation documented an open-source guide sitting at position 35 in search that still earned nine AI citations, because the page was built to be quoted. Being liftable and being top-ranked are two different competitions.
What is a "Key takeaways" box, exactly?
It is a short boxed summary at the very top of a page, holding two or three plain sentences that answer the question the page exists to answer. It reads as a complete answer with no reference to the surrounding text, which is what makes it easy for both a skimming reader and an AI tool to use. Nothing about it needs to be technical, and it does not need a plugin.
Is E-E-A-T a Google ranking factor?
No. E-E-A-T describes what Google's human quality raters look for when they assess search results. It is not a switch in the algorithm and it cannot be bought as a service. The underlying signals, a named expert author, honest dates and cited sources, are still worth adding, because they make your page more credible to readers and more quotable to AI tools.
Will a takeaways box guarantee that AI quotes me?
No, and be careful with anyone who says otherwise. No study isolates how much a takeaways box alone lifts your citation rate. What the evidence supports is that AI search retrieves passages, and that an answer-first opening makes a page easier to retrieve and cite. It improves your odds. It does not buy a result.
Make the good page you already wrote count
You are not starting from behind on this. The hard part, knowing your subject well enough to answer the question honestly, is already done. What is missing is the packaging, and the packaging takes an afternoon.
So take your most important page. Add the key takeaways box near the top. Add your name, the dates, and your sources. Then leave it alone and go and do the same to the next page.
If you would rather have someone else look at your pages and tell you which ones are worth fixing first, that is our kind of job. We are a digital marketing specialist, not a full-service agency, and quiet, specific problems like this one are the ones we most enjoy sorting out. Tell us what you sell and we will tell you where you stand.
Sources: click-through figures from the Pew Research Center, published 22 July 2025. Position and citation data from Foundation's Atlan E-E-A-T playbook, published 10 July 2026.
