How to Write Content That Answers Buyers and Earns AI Citations

Nearly two-thirds of Stacker’s 100 most-cited stories were explainers, according to the supplied summary of its content analysis. That doesn’t tell you how often its explainers failed to get cited.

That gap matters if you’re deciding what to write next. A list of winning pages can suggest useful ideas. It can’t prove that copying their format will make your page win.

For a small SaaS business, the sensible starting point is a buyer’s question you can answer with evidence. Formatting comes after you have something worth citing.

What is AI citations content?

Content earns an AI citation when an answer engine links to it as a source. Stacker’s analysis describes citations as links accompanying answers from tools such as ChatGPT and Gemini.

“AI citations content” isn’t a separate species of writing. It’s a loose label for pages you want those tools to use when answering relevant questions. For a SaaS founder, that might mean an explanation of a technical problem, an honest product comparison, or research you can show your work for.

The useful question isn’t whether you should write for humans or machines. Your buyer has a problem, and an answer engine needs sources that help answer it. Start where those needs overlap.

Suppose your product helps people move data between tools. A buyer might ask whether a move preserves their history. A useful article would explain what transfers, what doesn’t, and how you checked. That’s a proposed article, not a claim about what any particular engine will cite.

A page that merely repeats “data migration” throughout its headings hasn’t answered the question. Neither has a polished introduction followed by a sales pitch.

Write the answer you can defend before you optimise the page that holds it.

Citations also need to stay separate from business results. The studies discussed here examine which pages engines cite. They don’t establish that a citation will send you a buyer or produce a sale.

Choose questions connected to a real buying decision. Then a useful page still has a job if an AI engine never links to it: helping someone understand whether your product fits.

Matching the buyer’s question beats copying a page template

Two recent preprints found that pages matching users’ questions received more citations within the websites studied. Neither offers a reliable recipe for turning a template change into more citations.

Ben Moore and Liam Dunne’s preprint, submitted on 28 September 2026, examined approximately two million citations across 10,000 pages from nineteen B2B SaaS workspaces. Its strongest page-level predictor was overlap between page text and the collection of prompts tracked for each workspace.

That measure deserves care. The researchers measured shared words across a prompt collection. They didn’t test whether interviewing one customer and rewriting one article would increase citations. Still, the finding points toward a better starting question than “Should we add an FAQ?”

Ask what your intended reader actually wants answered.

Mahe Chen and Nick Koudas’s preprint, submitted on 10 October 2026, examined 158,099 answers containing 468,037 citations. Comparing pages on the same website, it found that matching titles and paragraphs to commercial questions mattered. FAQ blocks and several other familiar page features appeared effective only when comparing different websites.

For your next article, write down the question before choosing a format. Make it narrow enough that you can give a useful answer without writing a guide to your entire market.

“Which customer support tool should I buy?” is broad. “Can I move my existing support history without losing attachments?” gives you a clear job.

That example isn’t a tested citation tactic. It’s a practical way to apply the evidence without pretending the evidence promises an outcome. You can judge whether the finished page answers that question, even before you have citation data.

A cited page doesn’t prove its FAQ block worked

FAQ blocks can appear helpful when researchers compare websites, then lose that apparent benefit when researchers compare pages within the same website. Moore and Dunne report that pattern for FAQ blocks, structured data, and Core Web Vitals.

Their results changed after the analysis accounted for differences between domains. The positive relationships reversed or fell to zero. That’s a warning against treating the visible traits of cited pages as instructions.

Imagine a well-known website with strong reporting and carefully maintained pages. It also uses FAQs. A smaller website publishes thin answers without FAQs. Comparing those websites could make the FAQ look responsible for the gap, even if something else explains it.

That’s an illustration, not a finding about two named websites. The point is that pages arrive with baggage. Their domain, reporting, and reputation don’t disappear because someone puts their layouts in a spreadsheet.

Before spending a week on a checklist for generative engine optimisation, or GEO, ask three questions:

• Did the researchers change pages, or compare pages that already differed?
• Did they compare pages within the same website?
• Did they measure citations, or something else?

A checklist can still help you catch broken pages or organise a long answer. It just needs an honest label: publishing cleanup, rather than a proven route to citations.

If an FAQ answers a real follow-up question, keep it. If it repeats the article solely because a checklist demands one, ask what the reader gains.

A feature appearing on cited pages is evidence of a pattern, not proof that adding it caused the citations.

What does the explainer finding actually tell you?

Explainers made up nearly two-thirds of Stacker’s 100 most-cited stories, according to the supplied summary of its analysis. That describes the winning group, not the success rate of every explainer it published.

To estimate that rate, you would need to know how many explainers were available and how often they earned citations. You would also need comparable figures for other formats. The top 100 alone can’t answer those questions.

The article highlights four formats: explainers, rankings, research, and comparisons. Treat them as useful ways to organise an answer, rather than four buttons that make an engine cite you.

Choose an explainer when the buyer needs to understand how something works. Choose a comparison when they need help deciding between options. Publish research when you have evidence that adds something to the question.

For a tiny SaaS business, “research” doesn’t have to mean pretending you run a large survey company. You could document a product test, provided you clearly state what you tested and what the result does and doesn’t show. That’s an editorial suggestion, not a result established by these studies.

An honest comparison also needs room for your product to lose. If an alternative is better for a particular need, say so and explain why. The goal is a defensible answer, not a table where you award yourself every tick.

The analysis itself acknowledges that authority matters and that it can’t isolate the effect of framing and angles. Keep that caveat attached to the format advice.

Pick the format that lets you answer clearly. Don’t turn a description of successful stories into a promise about your next post.

How much can a better page change your citation chances?

Page features explained much less of the variation in citations than retrieval did in one recent study. Chen and Koudas report 32% for retrieval versus 2.8% for page properties in their analysis.

Retrieval means finding pages before writing an answer. In the researchers’ ChatGPT replay, every cited page had first appeared in the system’s own search results. The model rewrote questions into searches, and those searches rarely matched results from Google or Bing.

Those numbers don’t mean you can buy a 32% improvement by changing your content. They describe how much variation the researchers could explain in their data. They aren’t predicted gains for your website.

The distinction matters because writing a better page and getting that page found are related, but they aren’t identical jobs. A useful answer still has to enter the set of pages an engine considers.

The same preprint reports that model knowledge shaped searches and brand choices. It also describes controlled experiments in which brand names changed rankings. That is experimental evidence about brand names, not proof that adding an FAQ will produce more citations.

You don’t need to respond by stuffing brand names into every paragraph. The supplied evidence doesn’t support that advice.

Instead, set a limit on what you expect from an article rewrite. Improve the answer because you can inspect its accuracy and usefulness. Watch whether engines find and cite it, but don’t assume you control every step between publishing and appearing in an answer.

For a solo founder, that limit is useful. It stops a week of tiny formatting edits from becoming an endless attempt to control a system whose choices also depend on search and prior knowledge.

How should you write your next SaaS article?

Answer one buying question and attach evidence to the claims that carry the answer. The question-matching findings in Moore and Dunne’s study support that priority, though they don’t prove this workflow will increase citations.

Start with a question you’ve actually received. If you haven’t saved customer questions, choose one you can verify with a customer before drafting. Don’t invent demand because a writing tool produced an attractive outline.

Use a short working document with four parts:

1. The question. Write the buyer’s words, including any constraint that changes the answer.
2. The answer. State what you can honestly say before writing the introduction.
3. The evidence. Gather the documentation, tests, or sources behind the important claims.
4. The limits. Record what depends on the buyer’s setup and what you haven’t checked.

Then choose the format. A question about how something works probably needs an explanation. A choice between products needs a comparison. A claim about measured behaviour needs the underlying evidence and a clear account of how you got it.

Draft the answer near the top. Use headings that describe the actual sub-questions. Keep each section clear enough that a reader can understand it without hunting through the rest of the page.

Those are readability choices, not a proven citation formula. Don’t give them a stronger sales pitch than the research allows.

Before publishing, remove claims you can’t support. Check that links lead to the evidence you describe. Check that the article acknowledges the cases where your answer changes.

You now have a useful asset even if citation results disappoint: a page you can send to the next buyer who asks that question. That’s a better use of limited writing time than a generic article built around a formatting checklist.

How can you test changes without fooling yourself?

A before-and-after citation count can show a change without proving your edit caused it. Chen and Koudas describe engines rewriting searches for each answer, which gives you a concrete reason to expect the path to a citation to vary.

Keep a small, repeatable test instead of relying on a screenshot of one flattering answer. Choose questions tied to the article’s purpose and record exactly what you ask.

For each check, save:

• The question, engine, and date.
• Whether your page appeared as a linked source.
• Whether the answer used the page accurately.
• What you changed on the page between checks.

Keep brand mentions separate from linked citations. The preprint reports that a model can name a brand without a supporting page. Counting both as the same result would blur what you’re trying to learn.

If you change the title, rewrite the answer, add an FAQ, and publish new research together, you won’t know which change helped. You may still decide the rewrite was worth doing. Just don’t turn the result into a confident claim about one ingredient.

Where possible, leave comparable pages unchanged while testing an edit. That won’t remove every source of uncertainty, but it gives you more to inspect than a single before-and-after screenshot.

Also track the result your business needs separately. A citation is one observation. A useful conversation with a buyer is another. Don’t substitute the easier count for the outcome you actually care about.

For your next post, spend the first working session collecting the question and its evidence. Finish the draft before opening the GEO checklist. Then publish, record what happens, and keep the difference between “we observed this” and “we caused this” intact.