Product Pages Dominate AI Citations: What It Means for B2B

Product pages AI citation turns out to matter far more than the common assumption that Reddit and YouTube are the sources AI "prefers." In the Ten Speed study covered by Search Engine Journal, product pages accounted for 24.1% of all citations that appeared once a B2B buyer had already entered the vendor-evaluation stage.

This finding is summarized in a Search Engine Journal article on product pages earning 24% of AI citations. The study used 170 prompts and produced 7,387 total citation appearances across ChatGPT, Perplexity, Claude, and Gemini.

But this figure needs to be read carefully. The dataset only covers Ten Speed's own B2B SaaS and professional services clients plus look-alike competitors. The number of client brands isn't published, there's no per-platform breakdown, and the headline comparison of 24% versus 4% is descriptive, not the result of a dedicated statistical test.

What Did the Ten Speed Study Actually Measure?

Ten Speed used citation data from Peec AI, a tool that monitors which URLs get cited by ChatGPT, Perplexity, Claude, and Gemini.

They built 170 prompts designed to represent a buyer who has already moved past the category-education stage.

No longer:

"What is a CRM?"

but queries like:

"Pipedrive vs HubSpot for a sales-led company"
"How does X handle SOC 2 reporting?"

So the dataset sits much closer to bottom-of-funnel evaluation than top-of-funnel discovery.

How Much of the Citation Share Comes From Product Pages?

Product pages are the single largest category, at 24.1% of all citation appearances.

YouTube alone is only about 1%, according to the source article.

Why Do Product Pages Dominate Citation Share?

For a buyer already comparing vendors, a product page is often the most direct source to answer questions like:

• what does this product do;
• who is it for;
• what features are available;
• what integrations does it support;
• what use case does it fit;
• how does its pricing or packaging work.

In a B2B context, buyer evaluation usually needs specific evidence, not just generic social proof.

A product page becomes the natural place to find that evidence, provided it's written clearly.

What Does the 88.3% Brand-Controllable Content Figure Mean?

Ten Speed calculates that content a brand can control makes up 88.3% of all citations in the dataset.

This category covers pages written and owned by a marketing team, such as:

• product pages;
• comparison pages;
• articles;
• how-to guides;
• homepages.

This finding challenges the narrative that AI search mainly leans on third-party community content.

But again, the context here is B2B buyer evaluation, not every type of AI query.

Does This Mean Reddit Doesn't Matter?

No.

SEJ actually stresses that Reddit and YouTube advice isn't wrong — its relevance likely just differs by funnel stage.

Community content may matter more when a user is still asking:

"What's a good piece of software in this category?"
"What has people's experience been with tool X?"

While product pages become more relevant once the question shifts to:

"Does X support this specific requirement?"
"X vs Y, which one fits better?"

So source preference can shift based on intent.

Why Are Comparison Pages So Interesting?

Comparison prompts made up only around 20% of the prompt set, but produced almost 27% of the citations.

Ten Speed calls this roughly a 1.33x return relative to its share of prompts.

That means comparison content over-indexes relative to the volume of queries tested.

For B2B, pages like:

• X vs Y;
• X alternatives;
• X vs Y for a specific industry;
• X vs Y based on a specific feature;

can become a more important citation surface than an internal battlecard ever was.

Does the 1.33x Figure Prove Causation?

No.

That number shows overrepresentation within the dataset, not proof that building a comparison page automatically increases citations by 33%.

This is an important distinction.

Observed:
comparison prompts = 20%
citation share = ~27%

Not proven:
building a comparison page → citations rise 1.33x

Why Does the Homepage Still Get 7.8%?

A homepage is often assumed too generic for citation, but in this dataset its share is meaningful.

SEJ suggests writing the homepage and product pages so a model reading "cold" can understand:

• what the company does;
• who the target user is;
• what problem it solves;
• the product category;
• the main integrations or use cases.

Copy that's too abstract can force the system to infer far too much on its own.

What Does Writing "Like AI Reads It Cold" Actually Mean?

This doesn't mean writing for a robot.

It means a page should still be clear even when the reader has no prior context.

An example of copy that's too abstract:

"Transform your future with intelligent orchestration."

A clearer version:

"This platform helps IT teams
automate user provisioning,
access approval, and offboarding
across the company's SaaS applications."

The second version gives both humans and machines far more factual material to work with.

What Role Do G2 and Capterra Play?

Directory profiles like G2 and Capterra account for 7.2% of citations in the dataset.

That makes a directory profile far more than just a review inbox.

Information worth keeping current includes:

• category;
• description;
• integration list;
• feature tags;
• company details.

If a profile is stale or filed under the wrong category, AI can end up with an inaccurate representation while building a shortlist.

Is a Directory Profile More Important Than Reddit?

In this dataset, yes, in terms of citation share: 7.2% versus a combined 4.2% for Reddit, YouTube, forums, and discussions.

But that figure shouldn't be generalized to every vertical or every funnel stage.

The dataset is B2B SaaS and professional services.

What's the Biggest Limitation of This Study?

SEJ actively pressure-tested the study's methodology with six questions and found several important limits.

1. The number of brands isn't published

Ten Speed doesn't provide the number of client brands or a range, out of concern it could identify a small client base.

That makes it impossible for a reader to gauge how broad the sample actually is.

2. There's no breakdown per AI platform

ChatGPT, Perplexity, Claude, and Gemini are all combined.

Yet it's entirely plausible each one has different citation behavior.

3. The headline split wasn't tested statistically

The comparison of 24% product pages versus 4% community sources is reported descriptively.

Ten Speed does use a nonparametric test elsewhere, but not for this headline split.

4. It's a point-in-time dataset

This is a snapshot, not a longitudinal trend.

5. Citation isn't the same as pipeline

The study doesn't connect citation to clicks, demos, opportunities, or closed revenue.

Why 170 Prompts, Not 220?

SEJ found an inconsistency in Ten Speed's initial visuals.

One table listed 220 prompts, while another chart used 170.

After being asked, Ten Speed confirmed 170 is the correct figure and the incorrect visual would be fixed.

This matters because the denominator changes every single percentage.

Why Did This Methodology Actually Become More Trustworthy After Being Criticized?

SEJ's Greg Jarboe judges that Ten Speed's willingness to admit the data gap increases trust more than hiding it would have.

A study that says "we don't have this breakdown" is more useful than one that turns limited data into a universal benchmark.

A principle that can be applied to every AI citation study:

Ask about the denominator
Ask about the platform breakdown
Ask about the sample size
Ask about the statistical test
Ask about the business outcome

What Should Be Fixed on a Product Page?

If the target is bottom-of-funnel AI evaluation, a product page should answer the basics explicitly.

1. Explain the product in plain language

Don't just lean on a slogan.

2. State the target audience

For example:

"For IT teams with 500+ SaaS users"

3. Explain the capability

Use factual statements.

4. List the integrations

When integration is a major part of the buyer's decision.

5. Explain the limitations

Qualification also means knowing when a product isn't the right fit.

Why Can Stating Limitations Actually Help?

AI buyer evaluation isn't just looking for "fit" — it's also looking for "not a fit."

A page that's too promotional and never explains:

• plan limitations;
• deployment requirements;
• supported regions;
• minimum seat count;
• integration boundaries.

can push the system to go find the answer somewhere else.

How Do You Write a More Useful Comparison Page?

A comparison page shouldn't just state that your own product always wins.

A more credible structure:

A factual comparison gives a model far more material to work with than defensive copy does.

Does Article Content Still Matter?

Yes. Articles, blog posts, news, and PR contribute 17.4% of citations — the second-largest category after product pages.

That shows content marketing still plays a major role.

What's changing is resource allocation.

If the entire content budget only goes into top-of-funnel articles, a brand may end up neglecting the citation surface at the evaluation stage.

What's the Difference Between Top-of-Funnel and Bottom-of-Funnel AI Content?

Ten Speed's dataset is mostly talking about that third stage.

Why Does Funnel Stage Need to Be Tracked in AI Visibility?

Because the citation pattern can shift based on intent.

If a dashboard mixes together:

"what is a CRM?"
+
"HubSpot vs Pipedrive?"

the resulting source mix becomes far less meaningful.

A better segmentation:

• awareness prompts;
• category prompts;
• comparison prompts;
• feature qualification;
• purchase readiness.

Is Brand-Owned Content Always More Trusted?

No.

The study only shows brand-controllable content getting a large share of citations in this dataset.

It doesn't prove AI "trusts" brand-owned content more as a universal rule.

Product pages may get cited simply because they're the most direct source for product facts.

Trust, citation, and authority aren't the same thing.

What's the Connection to the Study Showing AI Recommends Brands but Cites Other Sites?

A different study SEJ has previously covered found owned brand pages contributed only a small share within a particular dataset, while third-party sources dominated citations.

The two results don't necessarily cancel each other out.

Possible reasons for the difference:

• different prompts;
• different verticals;
• different funnel stages;
• different platforms;
• different samples;
• different classification.

If your team has previously covered AI brand recommendation or citation sources, the internal article Google and ChatGPT Used Together: What Happens to Queries and Clicks? can serve as an internal link once the previous article's URL is verified.

How Do You Measure Product Page Citation?

Use prompts that genuinely require product facts.

1. Gather sales questions.
2. Pull high-intent queries from PPC.
3. Build comparison prompts.
4. Build qualification prompts.
5. Run them across AI platforms.
6. Log the cited URL.
7. Classify the page type.
8. Connect it to pipeline data where possible.

What Metric Matters More Than a Raw Citation Count?

A citation count is just one layer.

More useful metrics:

• citation share for high-intent prompts;
• brand mentions;
• recommendation inclusion;
• citation accuracy;
• AI referral traffic;
• demo source attribution;
• assisted pipeline;
• closed revenue.

Ten Speed itself acknowledges it doesn't yet have data connecting citation to pipeline.

What Should a B2B Content Team Actually Do?

A more balanced set of priorities:

1. Audit product pages.
2. Fix homepage positioning.
3. Build comparison content.
4. Keep directory profiles current.
5. Keep using community content for earlier-stage buyers.
6. Don't stake the whole budget on one study.

Should Budget Shift From Reddit to Product Pages?

Not automatically.

The decision should follow stage and audience.

If a target buyer is active on Reddit during the category-education stage, community content still matters.

If the biggest bottleneck sits in vendor evaluation, product and comparison content may deserve higher priority.

Use your own customer-journey data.

What Shouldn't Be Concluded From This Study?

• Product pages don't automatically get 24.1% of citations in every industry.
• Reddit isn't always just 4%.
• YouTube isn't always just 1%.
• There's no breakdown across ChatGPT, Claude, Gemini, and Perplexity.
• The headline split wasn't tested statistically.
• Citation isn't the same as clicks or revenue.
• The dataset doesn't represent consumer e-commerce.

FAQ About Product Pages and AI Citation

What share of citations do product pages get in the Ten Speed study?

Product pages contribute 24.1% of the 7,387 citation appearances analyzed.

What share do Reddit and YouTube get?

Reddit, YouTube, forums, and discussion threads combined make up about 4.2%. YouTube alone is about 1%, according to the SEJ article.

How many prompts were used?

The figure Ten Speed confirmed is 170 prompts. An earlier visual that stated 220 was declared incorrect.

Does the study compare each AI platform separately?

No. ChatGPT, Perplexity, Claude, and Gemini are combined, and per-platform data wasn't collected in this particular pull.

Are comparison pages effective?

Comparison prompts made up about 20% of the prompts but produced almost 27% of the citations, an over-index of roughly 1.33x within the dataset. This isn't proof of a causal lift.

Does a citation mean a lead?

No. Ten Speed states it doesn't yet have data connecting citation to clicks, demos, pipeline, or a closed deal.

Conclusion

Product pages AI citation is the most striking finding from the Ten Speed study: 24.1% of 7,387 citation appearances came from product pages, while Reddit, YouTube, forums, and discussion threads combined made up only 4.2%.

But context shapes what that number actually means. The prompts focus on B2B buyers already evaluating named vendors, not users still learning the category. The dataset also only covers Ten Speed's own client base and look-alike competitors, has no per-platform breakdown, and its headline split is descriptive.

For a B2B content team, the most useful takeaway isn't "abandon Reddit." The priority is making sure product pages, the homepage, comparison pages, and directory profiles genuinely answer evaluation questions in specific, factual language.

A healthy AI visibility strategy still needs measurement by funnel stage, by platform, and ultimately by business outcome — not just a raw citation count.

If your business wants to build AI citation monitoring, B2B product content, comparison architecture, or a generative AI visibility dashboard connected to buyer intent and pipeline, you can discuss your business's technology needs with our technical team.