Getting cited and recommended · · By Gerald De Oleo
The AEO Advantage Competitors Can't Copy: Your Product Data
The short answer: AI engines cite content that is specific, attributable and checkable. The most durable version of that is original data — a benchmark, a study or an aggregate from your own product that nobody else can produce. Every other tactic can be copied by a competitor next quarter. A number only you have can’t be.
Most B2B SaaS companies are sitting on this and publishing blog posts about best practices instead.
Why data gets cited
This isn’t a hunch. The research that coined the term GEO tested nine different ways of rewriting content and measured how much each changed that content’s visibility inside AI-generated answers. The strongest methods were adding citations, adding quotations and adding statistics — each improving visibility by roughly 30–40% on the paper’s main metric. Keyword stuffing performed worse than leaving the content alone.
I covered that study in more detail in AEO vs GEO. The mechanism is simple enough: a model assembling an answer needs claims it can attribute. A sentence with a number and a source is usable. “Industry leaders agree that alignment is critical” is not.
“Doesn’t that just mean citing other people’s statistics?”
That works, and you should do it. But it puts you one step down the chain — you’re the page that quoted the number, and the engine may cite the original instead.
Publishing the number puts you at the top of that chain. Everyone who wants to use it has to point at you.
The proof is this blog
Worth being concrete, because I’ve been doing exactly this while writing here.
Across the posts on this site I’ve cited Pew Research on click behaviour, Ahrefs on citation overlap, SE Ranking on review platforms, G2 on buyer behaviour, a Princeton-led research paper on generative engines, and Vercel on AI crawler traffic.
Every one of those organizations published a number nobody else had. Pew ran a panel study. Ahrefs queried its own index. Vercel measured traffic across its own network — data that exists because of what they operate, not because of what they researched.
None of them paid for those mentions. They published something measurable, and the citation is self-sustaining, because an article about AI crawler behaviour needs Vercel’s number.
That’s the asset. Your product generates equivalent data.
What you already have
Five places to look, in rough order of how quickly you can publish.
1. Aggregate product telemetry. What your users actually do, at scale, anonymized. How long a process takes before and after automation. Error rates. Time-to-value. If your product replaces a manual workflow, you can measure that workflow better than any analyst.
2. Benchmarks by segment. The same metric, sliced by company size, industry or region. Benchmarks are the most linkable form of data because readers want to compare themselves to their own segment, and that creates a reason for other articles to reference you repeatedly.
3. Aggregate customer outcomes. Not a case study — the distribution. “Across 340 accounts, median reduction was X, with the top quartile at Y.” That’s more credible than one logo’s success story, and far more citable.
4. Market data only your position reveals. Pricing you observe, integration adoption rates, which tools appear alongside yours, seasonal patterns. If you sit in the middle of a workflow, you can see the market’s shape.
5. A survey of your users. The slowest of the five and still worth it, especially for questions your telemetry can’t answer — why people do something, not just what they do.
“Our data isn’t big enough to be interesting.”
Smaller than you think is fine. A study of 200 accounts with a stated methodology is more citable than a vague claim about “thousands of customers.”
What matters is that the number is specific, the sample is disclosed, and the question is one people actually ask. A tight finding from a narrow dataset beats a broad claim with no method behind it — and being honest about the limits is part of why it gets trusted.
Picking the number
Not every metric is worth publishing. The ones that get cited share five traits.
It answers a question people already ask. Check your own AI answers and search data for the questions in your category that currently get answered with guesses. That’s where a real number lands hardest.
It’s specific enough to quote. “Teams save time” is unusable. “Median close time dropped from 6.2 days to 3.8” is a sentence a model can lift whole.
It’s comparable. Segmented so a reader can locate themselves in it.
It’s falsifiable. Stated method, sample size, time period. Numbers without methodology get ignored by careful writers, and careful writers are the ones whose articles get cited.
It’s repeatable. If you can publish it again next year, it becomes the reference series for your category. That’s the compounding version.
Publishing it so it actually gets used
The format matters more than people expect.
One dedicated page per study, not a section inside a longer post. It needs its own URL to be cited.
The headline number in the first two sentences. Don’t bury the finding under methodology. A model extracting the claim shouldn’t have to read four paragraphs to find it.
Methodology visible on the same page. Sample size, dates, how you measured, what you excluded. This is what separates a citable study from a marketing claim.
A table or chart in plain HTML. If the numbers live in an image, they’re invisible to the engines that don’t render or parse images well.
Permission to use it. State plainly that others may cite the data with attribution. People reuse numbers they’re sure they’re allowed to reuse.
Don’t gate it. This is the one that kills most B2B data programs. A PDF behind a form can’t be read by any crawler, can’t be cited by any engine, and can’t be quoted by any journalist. If the goal is citation, the gate defeats the entire purpose. Gate a deeper cut if you must — publish the headline findings openly.
Everything else about making pages readable to AI crawlers — server-rendered HTML, crawler access, structure — is in How to Appear in ChatGPT Answers.
Getting the number into other people’s articles
Publishing is half the work. The other half is making the number easy to find and reuse.
Tell the people who write about your category. Analysts, newsletter writers, journalists covering your space, and the people who publish the comparison guides in your category. A single specific finding is a far better pitch than a product announcement.
Reference it in your own writing. Every subsequent post that cites your own study reinforces the association between your domain and the number.
Answer questions with it where your buyers are. Where the data genuinely answers a question being asked — communities, forums, comment threads — the number does the work. Not as promotion; as the actual answer.
Repeat it annually. The second edition is when it stops being a post and becomes a reference.
That last point connects to something from the citation research: the sources AI engines draw from skew heavily toward pages that other pages already reference. Original data is one of the few reliable ways to enter that loop from outside.
What fails
Vanity metrics. Your ARR, your headcount, your funding. Nobody else’s article needs those numbers.
Unfalsifiable claims. “10x faster” with no method. Careful writers won’t touch it.
Gated PDFs. Covered above, and worth repeating because it’s the most common mistake.
Data with no segment. A single global average that nobody can locate themselves in.
One-off studies. A single publication gets a spike. A series becomes the reference.
Publishing before the entity is fixed. If sources across the web still describe your company by its previous positioning, a study strengthens the wrong association. Resolve the description first — that’s the order I argue for in How to Get AI Recommendations.
Quick answers
What kind of content gets cited by AI engines? Content that is specific, attributable and checkable. Research on generative engines found that adding statistics, citations and quotations were among the most effective changes, while keyword stuffing performed worse than doing nothing.
Where do I find original data for content? Inside your product. Aggregate telemetry, segment benchmarks, distributions of customer outcomes, market patterns visible from your position in the workflow, and surveys of your users.
How big does a dataset need to be? Smaller than most teams assume. A few hundred accounts with a stated methodology is more citable than a vague claim about thousands of customers. Disclosure matters more than scale.
Should I gate my research report? Not if you want it cited. Gated PDFs can’t be read by crawlers or quoted by writers. Publish the findings openly and gate a deeper cut if you need the leads.
How often should I publish original data? Annually, on the same metric, is the version that compounds. The second edition turns a post into a reference series.
If you want to know which questions in your category currently get answered without any real data — and which number of yours would land hardest — that’s part of what the audit surfaces. Before we talk, I run a free Shortlist Check — three or four of your buyers’ questions in one engine — so you can see whether the full audit is worth it. See how the audit works, or book a call.