How I Structure Content So AI Cites It

Most of the AI-search conversation I hear is about traffic. Clicks are down, AI answers are eating the results page, and everyone wants their old numbers back.

I think that’s the wrong scoreboard. When an AI assistant answers your buyer’s question, the question that matters is whose facts the answer carries. Whose framework gets named. Who gets linked when the buyer asks for a source.

That’s a citation game, and it runs on different levers than the ranking game did. I ran a diagnostic version of this argument in my two-minute AI positioning test. This piece is the prescription, and it’s built on the playbook I taught at a recent mastermind session for consultants.

What the numbers actually say

Start with the scoreboard problem, because the data is blunt.

Rand Fishkin’s SparkToro research, built on Similarweb clickstream data, found that 68% of U.S. Google searches in the first four months of 2026 ended without a click to anywhere. Not just without a click to your site. Without a click at all. The same research shows Google’s AI Overviews now appear on more than 20% of searches, and when one shows up, click-through rates drop by nearly 60%.

So yes, the traffic is going away, and it’s mostly not your fault. Here’s the half of the story the panic skips.

The visitors who still arrive from AI are worth more. Semrush’s study of AI search traffic found the average AI search visitor is 4.4 times as valuable as the average traditional organic visitor, measured by conversion rate. Adobe Digital Insights’ April 2026 Quarterly AI Traffic Report found AI-referred traffic to U.S. retailers grew roughly 400% year over year and converts about 42% better than non-AI traffic, with longer and deeper visits. And Kevin Indig’s Growth Memo analysis of Similarweb data found ChatGPT’s transactional traffic converting at 6.9% in the U.S. against Google’s 5.4%.

The mechanism behind that conversion gap is old. Blue links gave people choices they had to test themselves. Click, skim, judge, and if the page didn’t satisfy the search, bounce back and try the next result. SEO practitioners call that pogosticking, and it signals an unsatisfied searcher. Google denies using it directly as a ranking signal, though the DOJ antitrust trial confirmed click-satisfaction data feeds its Navboost system.

AI collapses that loop. The assistant does the pogosticking for you.

The result? Fewer visitors, better visitors, arriving closer to a decision. Which means your content’s new job isn’t winning the click. It’s being the source the machine trusts enough to quote.

The three lanes of AI-era visibility

I teach this as three lanes, because “SEO is dead” arguments usually collapse them into one.

• SEO is the rankings lane. Classic search visibility, still doing heavy lifting. Ahrefs found only 38% of AI Overview citations match the top 10 traditional results, but ranking well still correlates strongly with getting cited.
• AEO is the answers lane. Answer engine optimization: can your page directly answer a specific question in a form an engine can lift?
• GEO is the citations lane. Generative engine optimization: does your page carry a stat, framework, or claim distinctive enough that an AI wants to quote it and name you?

You don’t pick a lane. The same page can win all three, and the strongest pages do. What follows is how I build them.

The four moves I call SOME

When I audit a page for AI visibility, I run four moves on it. Summarize, Organize, Modularize, Emphasize. SOME, for short. Each move maps to something the machine needs before it will build with your page.

Summarize. Put the answer first.

AI systems reward pages that commit to an answer up front. A short summary at the top, the gist in the opening paragraph or a callout box, and if the headline asks a question, the answer in the first lines under it. This inverts how most experts write. We build context, then deliver the conclusion. Machines (and skimming buyers) read it the other way. The article you’re reading opens with a summary block for exactly this reason. The military has a name for this discipline: BLUF, bottom-line up front. Same skill, new reader.

Organize. Structure the evidence.

Give the machine something checkable to carry. Tables, charts, checklists, FAQs, and above all, claims with named sources linked to the original data. My standing rule: a statistic gets published with a named source or it doesn’t get published. Aggregator numbers with no traceable study get cut. That rule predates AI search. It just pays better now, because every unverifiable claim is a reason for the machine to build with someone else’s page. That includes my own numbers. First-party data needs a name and a date attached, or the machine can’t check it either.

Modularize. Make sections that stand alone.

AI systems lift passages, not pages. Clear section headers, anchors, a table of contents, and sections written so each one makes sense on its own, without the paragraph above it. A useful test: pull any H2 section out of your article and read it cold. If it depends on the three paragraphs before it to mean anything, it can’t be quoted. Rewrite it until it can travel alone.

Emphasize. Show the authority signals.

This is the trust layer. An author byline linked to a real bio. Credentials and firsthand experience made visible. Original research, data, or a case study. Reviews and social proof. A visible updated-on date. Schema markup so the machine can read who wrote what, and when. None of these are decoration. They’re the machine’s due-diligence checklist, and most expert content fails it while the experts wonder why thinner competitors get cited.

Beyond the page

SOME fixes pages. Four more moves work at the level of your whole presence, and they compound everything above. They’re not a framework, just a way I remember them: four Fs.

Fan-out. AI systems expand one question into a set: what is it, how does it work, what does it cost, what are the alternatives. A topic built as a hub with satellite articles answers the set. An isolated post answers one slice. It’s why my site runs on framework hubs like Power Positioning rather than scattered posts.

Fidelity. Your brand facts, exact and identical, on every surface machines cross-check. When your About page, LinkedIn, schema, bios, and book jacket all state the same facts in the same terms, the machine reads one confident entity. When they disagree, your presence in answers erodes. This is copy-paste faithfulness, not creative variation. Same title, same framework names, same numbers, everywhere.

Frameworks. Put your name on your thinking. Generic advice gets summarized without attribution because there’s nothing to attribute. A named framework has to be cited by name or lose its meaning. That’s the Multiply pillar working in a new arena, and it’s why this article teaches SOME rather than “four content tips.”

Footprint. The surfaces you don’t control. Reviews, community threads, podcasts, third-party mentions. Machines weight validation they can’t suspect you of writing. Off-page authority was already the hard part of organic visibility. AI search raised its price and its payoff at the same time.

What the winning sites share

Cyrus Shepard studied more than 400 websites that won or lost traffic through this shift, and the pattern in the winners (reported in the same SparkToro piece) is worth staring at. Winners offer a product or service, own proprietary assets, stay tightly niched, carry a recognizable brand, and, above all, let visitors complete a task. About 84% of winning sites let users finish what they came to do, against roughly half of the losers.

The traits also stack. Sites with one of them won about 15% of the time. Sites with four or five won around 68 to 70%.

Read that as a positioning finding, not an SEO finding. The sites surviving the zero-click era are destinations with a narrow focus, owned assets, and a name people search for directly. That’s Power Positioning described by someone else’s dataset.

Why AI can’t fake the inputs

Notice what everything above has in common. Verifiable evidence comes from work you actually did. Consistent facts come from a real history. Named frameworks come from original thinking. Third-party validation comes from relationships. None of that appears on demand.

That’s the point I keep making to leadership teams: AI doesn’t manufacture authority. It reads the signals that human authority produces, and it surfaces them.

Which is why the engine underneath this playbook is human. I call that layer EAT 2.0, my humanization frame: empathy, authenticity, and transparency. Empathy produces the buyer-specific depth that makes content worth citing. Authenticity produces the distinct voice that survives summarization. Transparency, about your methods, your numbers, and where AI fits in your own work, produces the trust signals machines and buyers both read.

Run SOME without that layer and you get well-structured content with nothing inside it. Run it with the layer and the citations compound.

Where to start

Don’t start with tools. Start with one page. Pick your most important article and score it against the four moves. Is the answer up front? Is the evidence structured and sourced? Can each section travel alone? Are the authority signals visible?

Then retrofit that one page before touching anything else. In the session where I teach this, that’s the whole homework: one page, four moves, thirty days. Most firms discover the problem isn’t AI at all. It’s that their expertise was never packaged specifically enough to be citable, by machines or by people.

That’s a positioning problem wearing a technology costume. It’s also the specific thing I diagnose in fractional engagements. Let’s have a chat.

---

Frequently Asked Questions