> ## Content Index
> Fetch the complete content index at: https://www.datadab.com/blog/llms.txt
> Use this file to discover other available public pages before exploring further.

# What answer engines cite is a listicle with a year in the title, not your customer story
- URL: https://www.datadab.com/blog/ai-cites-listicles-not-your-case-study/
- Published: 2026-11-03T04:30:00.000Z
- Updated: 2026-11-03T04:30:00.000Z
- Description: Your case study proves nothing to an answer engine, and that is not a fault. What the 328 cited URLs tell you to write instead.
- Author: Amit Ashwini
- Tags: AI Citations

*We measured 100 prompts against ChatGPT and 821 published case studies. The two datasets barely overlap, and that tells you which page to write next.*

Every content team I meet has a story hub and a suspicion: that the work they do to prove results is not the work that gets them named in an AI answer.

The suspicion is right, and the reason is not a conspiracy. It is format.

We ran [a study of what ChatGPT actually cites](https://www.datadab.com/research/what-chatgpt-actually-cites) across 100 prompts, and separately coded [821 customer stories from 479 B2B software companies](https://www.datadab.com/research/b2b-saas-case-study-index). Put the two side by side and the mismatch is obvious.

One of these assets is in the retrieval set. The other is not.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-01-hero-1.png)

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-02-the-cited-set-has-a-shape-and-yo-1.png)

## The cited set has a shape, and you can describe it in one line

Across 328 cited URLs, 134 of them (41%) carry "best" or "top" in the URL path. Eighty-two (25%) contain "comparison", "vs" or "compare". A further 159 of 328 titles (48%) include a year.

Read those three numbers together and the archetype is unmistakable: a dated roundup that compares things, written by someone who is not the vendor.

That is not an accident of our sample. It is what an answer engine is built to do.

It retrieves a passage that answers a question in the shape of a question, and a page titled "Best CRM for a 10-Person Sales Team (2026)" is closer to a question than anything your marketing team publishes.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-03-blog-content-beats-documentation-1.png)

## Blog content beats documentation ten to one

133 citations pointed at blog or article URLs. Five pointed at documentation.

That is a ten to one ratio, and it is the number I would put in front of a product marketing team that believes its docs are its best AI asset. They are not.

The engine is looking for prose that answers a buyer's phrasing, not reference material that describes a feature.

Documentation still earns the click. It just does not earn the citation. Those are different jobs, and conflating them is why teams keep building the wrong page.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-04-your-story-hub-is-a-first-party--1.png)

## Your story hub is a first-party asset in a third-party game

Here is what the 821 stories look like when you code them. The median story runs 921 words.

The result sits a median 7.8% down the page text, so the answer is up where a retriever can reach it. 42.0% of stories lead with a percentage headline, which works out at 60.5% of the ones that carry any metric at all.

Video appears in 21.4%, and a stated time-to-result in 11.3%.

That is a well-built asset. It is also, structurally, a vendor speaking about itself, and TechRadar at 23.0% of responses tells you which voice the engine prefers.

The top 10 domains take only 64 of 328 citations (20%), and 229 domains were cited exactly once, so the set is wide open.

It is still third-party editorial at one end and vendor marketing at the other.

I want to be careful with the claim here, because our two studies do not measure the same universe. The citation study counted what ChatGPT cited in answer to 100 category prompts.

The story study coded what vendors publish on their own domains. Neither counts "how often case studies are cited", so I cannot hand you that number.

What both support is narrower and still useful: the cited format is a comparison or roundup with a date in it, and a case study is neither of those things.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-05-the-power-law-is-steep-and-the-d-1.png)

## The power law is steep, and the door is not locked

Twenty percent of all citations go to the top ten domains. That sounds discouraging until you notice the other side of the distribution: 78% of cited domains appear exactly once.

Retrieval is not a shortlist of ten incumbents. A page that answers a specific question in a specific shape can enter the set on the strength of the answer.

That is the single most actionable finding in the whole panel, and it is why I argue against buying a sponsorship instead of writing the page.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-06-so-should-you-stop-publishing-ca-1.png)

## So should you stop publishing case studies

No. That would be the wrong conclusion from a real finding, and I have watched teams take it.

Customer stories are the human-evaluation asset. They exist because a shortlist needs a peer like you, with a problem like yours, at a size you can recognise.

The census shows that work scales with size: 33.3% of companies at 1-10 employees run a story hub against 90.2% of those with 200+.

Also, a third of published stories prove nothing at all. 251 of 821 (30.6%) carry no quantified result, which makes them testimonials with logos.

If you are going to publish a story, at least make it one a buyer can use, with the number and the timeframe in it.

Engines do not cite them. Buyers need them.

Keep both facts in your head at the same time and the plan writes itself: fix the stories for the human, and write the listicle for the machine.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-07-the-page-to-write-this-month-1.png)

## The page to write this month

Not another post. A specific artefact, with a specific shape.

Write the comparison or roundup that your category currently lacks, on your own domain, and give it the properties the cited set has. A date in the title.

Named alternatives, including at least one you do not win against. A table. A FAQ block answering the follow-up questions. Numbers attached to sources, with a verification date.

The honest version of this page is a competitive asset, and it only works if it is true.

A comparison that names every rival as a straw man gets skimmed by buyers and is not much better at retrieval than the story hub, because the shape is right but the substance is thin.

Two of our own studies carry those datasets. Cite the numbers you take, and be ready for someone to check them, because in this category the checking is now part of the job.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-08-the-cited-format-in-one-table-1.png)

## The cited format, in one table

Everything the panel says about shape, in one place.

| Property of a cited page                   | Count in the panel | Share          |
| ------------------------------------------ | ------------------ | -------------- |
| "best" or "top" in the URL path            | 134 of 328         | 41%            |
| "comparison", "vs" or "compare" in the URL | 82 of 328          | 25%            |
| A year in the title                        | 159 of 328         | 48%            |
| Blog or article content rather than docs   | 133 vs 5           | ten to one     |
| Citations going to the top ten domains     | 64 of 328          | 20%            |
| Domains cited exactly once                 | 229                | 78% of the set |

Hold the last two rows together and the picture is coherent: a steep head, a very long tail, and a format that is a dated comparison.

Nothing in that column is about word count, and nothing is about the author's job title. It is about shape.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-09-what-a-story-should-carry-if-you-1.png)

## What a story should carry if you are publishing one

Since the story hub is doing a human job, let it do that job properly.

The coded pattern across 821 stories is that the useful ones state a metric, a period and a buyer-sized context, and the weak ones state an adjective.

251 of the 821 (30.6%) carry no quantified result at all. Only 11.3% state a time-to-result, which is the single number a buyer actually wants: not that something improved, but over what window.

Video appears in 21.4%, which is a scale good rather than a proof requirement.

And anonymity is cheaper than people think. Anonymous stories quantify anyway, in our sample, which means the objection to publishing a story is usually about the logo, not about the number.

Get the number, lose the logo if you must, and stop pretending the second one is the barrier.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-10-so-what-should-the-page-be-calle-1.png)

## So what should the page be called

Copy the shape, not the competitor. A title with a year in it, a named use case, and a qualifier for who it is for is what the cited set actually looks like: 159 of 328 titles carry a year, and TechRadar reached 23.0% of responses across 8 of the 10 categories partly because it publishes in exactly that form.

The honest version of that sentence has a spine to it. A dated roundup can only claim a date, not permanence, so say what was verified and when, and let the date be the reason someone trusts you rather than the reason they distrust you.

If the number makes you uneasy, use it anyway with the caveat attached. The alternative is a page titled like a manifesto, which no engine can turn into an answer.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-11-who-this-is-for-1.png)

## Who this is for

\* *Content leads with a full blog and no citation:*\* stop adding posts, start adding one dated comparison per category you actually sell into. The panel says format beats volume.

\* *Founders who measure content by story output:*\* your stories are working, just not on the machine leg. Keep them, fix the third of them that carry no number, and add the roundup.

\* *Product marketers who believe docs are the AI play:*\* ten to one against you. Docs earn clicks. Editorial-shaped pages earn citations.

\* *Buyers:*\* this is why the "best X for Y" article you found via ChatGPT may be a vendor's own page wearing editorial clothes. Check who wrote it and when.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-12-a-note-on-how-confident-we-are-1.png)

## A note on how confident we are

The citation panel is 100 prompts against one engine at one moment. It describes that panel, not all ChatGPT behaviour, and answers move when models and retrieval indexes change.

Treat 2026 dates in it as a shelf life, not a constant.

The story index covers 821 stories from 271 of 479 companies that publish a hub at all, all YC-backed B2B software. Venture-backed bias, and the claims inside those stories are self-reported.

I describe presentation patterns, not the truth of a result.

The two datasets are not joined. No prompt in the citation panel was tested against a case-study URL, so "case studies are rarely cited" is an inference from format, not a measured rate.

It is the honest limit of this whole argument, and it is the reason the recommendation is about shape rather than a percentage.

![](https://www.datadab.com/blog/content/images/2026/10/ai-cites-listicles-not-your-case-study-13-wrap-up-1.png)

## Wrap-up

Dated comparisons and roundups are the cited format. Blog content outran documentation ten to one, a year appears in 48% of cited titles, and 229 domains entered the set once and never came back.

Your customer stories still matter, to the human half of the purchase. Write the roundup for the machine half, and [give the buyer the answer first](https://www.datadab.com/tools/answer-blueprint).

Citation figures come from 100 ChatGPT prompts and 328 cited URLs. Story figures come from 821 published case studies across 479 companies. The two datasets are not joined, and the post says so.

[Write the answer-first page](https://www.datadab.com/tools/answer-blueprint)