Content That Survives AI Overviews: 6 Formats That Still Get Clicks

The pages that got destroyed all had something in common.

They answered a question completely, in a way anyone could have answered it, and offered nothing else. "What is a DMARC record." "How to boil an egg." "Best time to post on Instagram." A user asks, an AI answers, and there was never any reason to click through — because the page contained only the answer, and the answer is now free.

The pages that held up had something the answer couldn't replace.

That's the whole principle, and everything below is an application of it. If a complete summary of your page satisfies the reader entirely, you've built something AI can replace. If the summary makes them want the page, you haven't.

Here are the six formats where that's reliably true.

1 — Original data and research

The strongest format, and it isn't close.

Original data is the one thing a synthesiser genuinely cannot produce. A model can rephrase every existing explanation of a topic; it cannot survey your 200 customers, audit your 40 client accounts, or tell anyone what your own conversion rate was last quarter.

This is why "add statistics" keeps appearing at the top of every citation study — Princeton's GEO research found statistics addition among the strongest of nine tested methods. But the usual reading of that finding is wrong. It isn't that models like numbers as a decorative genre. It's that a specific verifiable figure that exists nowhere else requires a source, and being that source is the entire game.

Analysis of what gets pulled verbatim backs this up: numeric formats — statistic lines and table rows — are copied most often, because numbers can't be safely reworded. Prose gets paraphrased; a number travels with its attribution.

What this looks like if you're small: you don't need a research department. You need to count something. Audit 30 sites in your niche and publish what you found. Survey your email list. Track your own results across a year and share the numbers. Aggregate what you've learned from client work into something quantified.

The bar is lower than people think. It just has to be yours.

2 — Interactive tools and calculators

An AI can describe how a calculation works. It can't be the calculator.

Anything requiring input, producing personalised output, or maintaining state is structurally click-necessary. A revenue calculator, an audit scorer, a checker, a template generator — the user has to arrive to use it.

These also tend to earn links naturally, which feeds the ranking that feeds the retrieval. And they convert better than articles, because someone who put their own numbers into your tool is now emotionally invested in the result.

The catch is that they cost more to build than a blog post. But one good tool typically outlives fifty posts, and it doesn't decay when the next model release absorbs your topic.

3 — Genuine opinion and analysis

Models are trained toward consensus. They're built to give you the settled, balanced, generally-accepted answer — which makes a strongly-held argued position something they'll represent cautiously and hedge around.

Which means "here's what I think, and here's why the common advice is wrong" is a format with structural protection. Not contrarianism for its own sake, which is easily seen through, but a real position you can defend from experience.

There's a second-order effect worth noticing too. Definitive language correlates with citation — pages that commit to a position get quoted more than pages that hedge. Having an actual opinion makes you both harder to replace and easier to quote. Those usually trade off against each other; here they don't.

4 — First-hand experience

The thing a model has no access to: what happened to you.

The client project that went sideways and what it cost. The tactic that works in your market but fails everywhere else. The mistake you made in 2023. The specific, situated, non-generalisable knowledge that comes from having actually done the thing.

This is also what E-E-A-T's first E is pointing at, and it's the one signal a well-resourced content team genuinely can't manufacture. They can out-publish you. They can't have been there.

For a personal-brand site this is the whole structural advantage. Every piece where you write "here's what happened when I did this" is a piece that only you can write.

5 — Comparisons and decision support

"Which of these should I choose" is a different question from "what is this," and it survives differently.

An AI answer can summarise the options. It's much weaker at the messy conditional reasoning a real decision needs — the tradeoffs that depend on your situation, the thing that looks best on paper but fails for a specific type of user, the "it depends, and here's exactly what it depends on."

Comparison content also sits closer to a purchase, which means it holds commercial value even at lower traffic. A comparison page getting a fifth of its old traffic but the same conversions is not really a loss.

Worth knowing: commercial-intent prompts trigger live web search far more often than informational ones do. Which means comparison and decision content is disproportionately likely to be retrieved fresh rather than answered from training — you're competing for a live citation rather than against a memorised summary.

6 — Deep, structured reference

The format people assume is dead, and isn't — with one condition.

Shallow how-to content genuinely is being eaten. A 900-word "how to set up DMARC" is exactly what an AI Overview replaces. But a comprehensive, well-structured reference that someone will bookmark, return to, and work through over an hour is a different thing entirely.

The distinction is whether the summary is a substitute or an advertisement. A summary of a thin post substitutes for it. A summary of a genuinely deep guide tells the reader there's an hour of useful material here and sends them to it.

Comprehensive guides with data tables show among the highest citation rates in the published research — and the citation itself functions as a recommendation.

The failure mode is fake depth: 4,000 words of padding around 600 words of substance. That gets summarised away like anything else, and now you've spent four times the effort for it.

What died, and why it's worth saying plainly

The formats that lost hardest:

Definitional content. "What is X" pages that only define X.

Simple factual lookups. Conversions, dates, single-number answers.

Thin how-tos. Procedures explainable in a paragraph.

Aggregated listicles with no original assessment. "10 best tools" assembled from other people's reviews.

Rewritten news. Coverage that adds nothing to the original report.

What unites them is that each was a summary of publicly available information. That was always a fragile business, and it's now largely an obsolete one. AI didn't invent that fragility — it just removed the friction that was propping it up.

The test to apply to anything you write

Before publishing, ask: could an AI answer produce a satisfying substitute for this page?

If yes, the page is at risk regardless of how well it's optimised.

Then ask the sharper follow-up: if someone read a perfect summary of this page, would they still want to visit it?

• For a definition, no. The summary was the value.
• For a calculator, yes. They need to use it.
• For original data, yes. They want the full dataset, the method, the caveats.
• For a strong opinion, often yes. They want the argument, not the position.
• For deep reference, yes. They want the material, not the map.

That question is a better editorial filter than any keyword tool.

Frequently asked questions

What content does AI not replace?
Content offering something a synthesiser can't produce: original data you collected, interactive tools requiring input, genuine first-hand experience, strongly argued opinion, and comprehensive reference material too substantial to summarise usefully. What AI does replace is content that summarises publicly available information — which is what most definitional and thin how-to content was.

Are listicles dead?
Aggregated listicles are, largely — "10 best tools" assembled from other people's reviews adds nothing a model can't assemble itself. Listicles built on original assessment are fine: if you actually tested the ten tools, that testing is the value and the list format is incidental. The format was never the problem; the absence of first-hand input was.

Should I stop writing how-to posts?
No, but raise the bar. Thin procedural content explainable in a paragraph is exactly what AI Overviews absorb. Comprehensive how-to guides with original screenshots, real troubleshooting from actual experience, and structured depth still earn visits — and comprehensive guides with data tables show among the highest citation rates in published research. The test is whether a summary substitutes for your guide or advertises it.

What to do next

Take your ten highest-traffic pages and sort them into two piles: pages where a perfect summary would satisfy the reader, and pages where it wouldn't.

The first pile is your exposure. Some of those pages can be upgraded — add original data, add a tool, add your actual opinion. Some can't, and it's worth knowing which.

Then look at what you have that nobody else does. Client results, your own numbers, an experience you haven't written up. That's your next piece, and it's the one that can't be replaced.

Free: The SEO audit checklist.

---

Related guides

• SEO when AI answers the question first — the wider strategy
• Entity SEO — making content specific enough to quote
• How to refresh old blog posts — upgrading the at-risk pile
• 7 linkable assets that earn links — tools and data as link magnets