Home / Blog / Getting Found When AI Answers
Day 23Getting Found When AI Answers the Question
The search engine used to hand people a list of doors. Now it walks in, takes what it needs, and comes back out alone.
Your content strategy was built on a trade: you write something useful, a search engine ranks it, and some percentage of readers arrive on your site where you can make your case. That trade is being renegotiated without your involvement.
Pew Research Center analysed real browsing behaviour in 2025 and found that when a search results page included an AI summary, users clicked through to any website on about 8 percent of visits. Without a summary, that figure was about 15 percent. They clicked a source link inside the summary itself on roughly 1 percent of visits. Half the traffic, gone, from the same query, on the same page, for the same content.
You are no longer competing for the click. You are competing to be the source the answer is made of.
Gartner has predicted a substantial fall in traditional search volume as assistants absorb informational queries. Whatever the exact figure turns out to be, the direction is not in dispute, and a startup whose growth model assumes today's click rate is planning with a number that is actively declining.
What actually changes
Three shifts, and each one has a practical consequence.
Being cited is the new being ranked. When a model composes an answer and names two sources, those two brands are recommended by something the reader trusts, without a click. That is closer to a referral than to a search result, and it converts better when the click does happen.
Volume falls, intent rises. The queries that vanish are the simple lookups: definitions, conversions, quick facts. Those visitors rarely became customers. The visitors who still arrive have made a decision to leave the answer and come to you, which makes them more valuable per head. Expect traffic down and conversion rate up, and update your reporting before someone panics at the traffic chart.
Structure beats volume. Publishing forty thin articles a month was already weak. Now it is worthless, because a model synthesising an answer has no reason to reach for the fortieth mediocre restatement of a common fact. It reaches for the page with a specific number, a named source and a clear claim.
Pew Research Center's 2025 analysis of real browsing behaviour. Click-through roughly halves when an AI summary is present, and users were also more likely to end their browsing session on that page. The traffic did not move to a competitor. It stopped existing.
How to be the source
Everything below is checkable and none of it requires a budget.
One question per page, answered in the first forty words
Open with the direct answer in plain language, then explain. A page that spends four paragraphs setting up context before answering is a page that gets skipped by readers and by anything summarising it. This is also better writing, which is a convenient coincidence.
Headings phrased the way people ask
"What is a good LTV to CAC ratio" outperforms "Understanding customer economics." The second sounds more considered and matches nothing anybody types or says out loud.
Specific numbers with named sources
"Retention improves profit" is unquotable. "A 5 percentage point increase in retention raised profits 25 to 95 percent in Bain's research" is quotable, checkable, and attributable. Systems selecting between sources favour material that can be verified, and so do humans.
Structured data that matches the visible page
FAQPage, Article and BreadcrumbList JSON-LD tell a machine what the page contains without inference. The rule is that the structured data must match what a reader sees, word for word where possible. Marking up questions that do not appear on the page is a fast route to being ignored or penalised.
Readable without scripting
If your content only appears after JavaScript runs, you are betting that every crawler and retrieval system executes it properly. Some do, some do not, and the ones that do not simply see an empty page. Server-render the text.
Entity clarity
Say who you are, consistently, everywhere. Same company name, same author name, same description on your site, your profiles and anywhere you are quoted. Systems that assemble answers connect entities across sources, and a company that describes itself three different ways is three weak entities instead of one strong one.
0 to 2. Invisible to answer engines. Start with items one and four on your five most important pages this week. 3 to 5. Partly ready. The gap is usually original material: you are restating what already exists. 6 to 7. Well structured. Now build the thing only you can publish, because structure without substance still loses. 8 of 8. You are a citable source. Keep measuring what assistants say about your category, monthly.
Eight checks, all verifiable in an afternoon. Item seven is the one that decides outcomes: structure makes you readable, but only original material makes you worth quoting.
Publish what nobody else can
The strongest position is having something no synthesis can reproduce, because it exists only in your business.
Original data qualifies. If you process ten thousand transactions a month in a niche, you can publish a quarterly benchmark nobody else can write. Named case studies with real numbers qualify. A method you have run a hundred times, written down step by step, qualifies. So does a teardown of your own failure with the arithmetic included, which almost nobody publishes and everybody reads.
General explanation does not qualify. "What is customer acquisition cost" is answered adequately by ten thousand pages, and yours will not be the eleventh that gets cited. Write the version that only you could write: your CAC by channel, over two years, with what you changed and what happened.
Open three different AI assistants. Ask each one the five questions your customers ask before they buy: "what is the best tool for X," "how do I solve Y," "who are the main providers of Z in my country," "what does X cost," "is X worth it." Record what each says, which companies get named, and what sources are cited. That transcript is your current market position in the channel that is replacing search, and most founders have never looked at it. Then read the pages that did get cited. They will be more specific, more numerate and better structured than yours, and now you know exactly what to write next. Repeat monthly and track whether you start appearing.
The channels that do not depend on being crawled
Answer engine work is worth doing and it is still a dependency on somebody else's system. Balance it with discovery routes nobody can restructure without asking you.
Email you own. A list of people who asked to hear from you is the only audience that cannot be reallocated by an algorithm change. It is also the reason to put a real reason to subscribe on every page.
Communities where your buyers already gather. Slack groups, industry forums, trade associations. Being genuinely useful there produces the referrals that AI summaries cannot intercept, because the recommendation comes from a person.
Partnerships and integrations. Being listed in someone else's marketplace or documentation puts you in front of qualified users at the moment of need, and it survives every search change.
Speaking and writing where the audience is assembled. A conference talk, a podcast, a guest piece in the publication your segment reads. Slower, harder to measure, and the resulting relationships outlast the platform.
How much control you hold over each route. The point is not to abandon the bottom two, which still work. It is to notice that a company whose entire acquisition sits in the bottom two has outsourced its survival to product decisions made by companies that have never heard of it.
Measure it differently
Sessions from organic search will fall even when your position improves, so a dashboard built on traffic will report a fire that is not burning. Change what you watch.
Track the share of assistant answers in your category that name you, sampled monthly against a fixed question list. Track branded search volume, which rises when people encounter your name inside answers and then look you up directly. Track conversion rate per visitor rather than total visitors, since the remaining traffic should be better qualified. And track direct and referral traffic, which is where citation-driven visits often land.
Those four together tell you whether the strategy is working. Total sessions on their own will tell you a story about decline that is mostly about the industry, and it will push you toward publishing more thin pages, which is the exact wrong response.
What this does not change
Two things survive every shift in how discovery works, and both are worth saying because founders in a panic tend to abandon them.
Trust still transfers through people. A recommendation from a peer in the same job outperforms any answer generated by any system, and it always has. The practical implication is that your customer relationships are a discovery channel, not just a revenue line, and asking a happy customer to name you in the forum where their peers ask questions is a five-minute action with a longer half-life than a year of content production.
And the buyer still has to choose. An answer engine can tell someone that three companies solve their problem. It cannot sit through the security review, absorb the risk of the decision, or agree to a start date. The last mile of a purchase is still a human being deciding to trust a specific provider, which is why the work in Day 12 and Day 15 keeps paying regardless of how the top of the funnel is reorganised.
The takeaway
- Clicks roughly halve when an AI summary appears, from about 15 percent to 8 percent. Plan for it now.
- Being cited replaces being ranked. Write pages a model can quote accurately.
- One question per page, answered in the first forty words, with numbers and named sources.
- Structured data must match the visible copy, and the copy must render without scripting.
- Publish what only you can: your own data, methods and teardowns. General explainers are already commodity.
- Build owned channels alongside, and measure citations and branded search rather than raw sessions.
Frequently asked questions
What is answer engine optimisation?
Structuring content so AI systems that answer questions directly use your material as a source and name you. It overlaps with classic search work, but the goal is being the source an answer is built from rather than winning a click from a list.
How much do AI summaries reduce clicks?
Pew Research Center's 2025 analysis of real browsing found clicks on about 8 percent of visits where an AI summary appeared, against about 15 percent without one, and source links inside summaries clicked on roughly 1 percent of visits.
How do I make my content easier for AI to cite?
Answer one specific question per page in the first forty words, use headings phrased as real questions, attach numbers to named sources, add FAQPage and Article structured data matching the visible text, and make sure everything renders without scripting.
Is SEO dead because of AI search?
No, but the value of a ranking is shifting. The queries disappearing are simple lookups that converted poorly anyway. Commercial and comparison queries still send people to sites. Keep earning citations while reducing dependence on any single discovery channel.
The traffic halved and the strategy did not change.
Kill My Startup is about the slow structural shifts founders keep treating as temporary.
Buy on Amazon →Sources
- Pew Research Center, July 2025, "Google users are less likely to click on links when an AI summary appears in the results," based on tracked browsing across 68,879 search queries in March 2025.
- Gartner press release on projected decline in traditional search engine volume as assistants absorb queries.
- Schema.org documentation for FAQPage, Article and BreadcrumbList structured data.
- Google Search Central guidance on structured data matching visible page content.