Missing from the answer
Six enterprise technology and services categories, UK, September and October 2026.
The findings
Trade media got between 3% and 9% of the sources AI cited, in every one of the six categories.
Vendor and company pages took between 52% and 80%. Review and analyst sites were cited more often than trade media in all six.
Across 783 answers to buying questions, a trade or business title was named in the answer text once. Gartner was named in 27.
In three categories, ChatGPT cited no trade media at all: not once in 998 citations.
Trade media was thin on Google’s first page too, at between 0% and 10% of results in two pulls a week apart. AI didn’t create the gap.
44 of 46 trade titles leave the door open, in their robots.txt, to the bots that fetch pages for live answers.
The published research on AI search says AI favours independent, impartial sources. Trade media is about as independent as it gets. So why are the publishers I talk to still watching their traffic fall?
The usual explanation is that nobody clicks inside AI answers. But if AI were building its answers from trade media, some of that would still show up. Having run publishers myself at Dennis and Autovia, I wanted to know why it didn’t add up, so I measured it.
This page carries the full data and method behind Issue 4 of The Selection Layer.
What I measured
Four things, in the same six categories: customer data platforms, HR and payroll software, autonomous financial services (AI agents in banking and insurance), identity governance, endpoint management, and SAP SuccessFactors implementation partners.
First, the sources AI cites when buyers ask buying questions, across ChatGPT, Gemini and Google’s AI Overviews. Second, which publications AI names in the text of its answers. Third, what Google’s own first page shows for the same questions. Fourth, how the engines differ from each other. The full method and its limits are at the end.
The Source Layer: what AI cites

Trade media got between 3% and 9% of the sources AI cited. Fewer than one in ten, in every category.
Most of the rest belonged to the vendors. Their own pages, and those of the consultancies, systems integrators and partners around them, took between 52% and 80%. Next came review and analyst sites, which AI cited more often than trade media in all six categories.
That last point matters more than it looks. Review platforms read like the independent voice in an answer, but many are paid by the vendors they list. Trade media is the independent voice, and it’s the one that’s largely missing.
Category | Google first page: trade media, 10 Oct | Google first page: trade media, 17 Oct | AI answers: trade media share of citations | Google first page: vendor and company sites, |
|---|---|---|---|---|
Customer data platforms | 4.9% (3 of 61) | 4.3% (3 of 70) | 2.9% | 77.0% / 77.1% |
HR and payroll software | 7.1% (5 of 70) | 5.6% (4 of 71) | 8.7% | 65.7% / 80.3% |
Autonomous financial services | 9.7% (16 of 165) | 4.6% (8 of 174) | 4.1% | 63.0% / 71.8% |
Identity governance | 0.0% (0 of 54) | 5.6% (3 of 54) | 3.6% | 59.3% / 63.0% |
Endpoint management | 2.4% (1 of 42) | 3.8% (2 of 53) | 2.6% | 73.8% / 56.6% |
SAP SuccessFactors partners | 2.0% (1 of 51) | 3.8% (2 of 53) | 2.6% | 78.4% / 69.8% |
Table 1. Trade media on Google’s first page and in AI answers, by category. Google: keyword versions of the buying questions, UK, pulled on 10 and 17 October 2026. AI: ChatGPT, Gemini and Google AI Overviews, September and October 2026. Categories in a fixed order, not a ranking.
Cited versus named
A citation is a source the answer shows as its evidence, usually as a link or a source card beside the text. A mention is a name in the answer itself.
You could argue the mention matters more. Buyers rarely click the links, but they do read the names, and a name in an answer is the closest thing AI has to an endorsement. So I counted those too, this time counting answers rather than sources. Across 783 answers to buying questions, a trade or business title was named in the text once: Reuters, in a financial services answer. Gartner was named in 27.
So trade media isn’t on the page either way. AI doesn’t lean on it for evidence, and it doesn’t name it as an authority. When an answer does name a reference, it names an analyst.
It names first and checks second
ChatGPT shows some of its working, which helps explain why we’re seeing this. In 36 of 59 answers, its very first search already named vendors.
So where do the names come from? From what the model already knows: everything it absorbed in training, long before anyone asked the question. Nobody outside the AI companies can see inside that. What we can see is what it does next, which is check its candidates, mostly on the vendors’ own pages.
That gives a publisher two places to be: in the model’s memory, and in what it checks. Trade media may well help shape what the model knows, but if it does, it isn’t getting the credit. The check is where the answer gets its evidence, and that’s where trade media is missing.
In three categories, ChatGPT cited no trade media at all, not once in 998 citations, even when it had pulled trade pages into its sources. In endpoint management, one trade title’s vendor ranking was cited eight times by Google’s AI Overviews. ChatGPT pulled the same page into its sources and cited it zero times.
Category | Google first page | ChatGPT | Google AI Overviews | Gemini | All three |
|---|---|---|---|---|---|
Identity governance | 0.0% / 5.6% | 0.0% (0 of 278) | 6.2% (29 of 470) | 2.2% (3 of 134) | 3.6% |
Endpoint management | 2.4% / 3.8% | 0.0% (0 of 333) | 7.2% (19 of 265) | 0.6% (1 of 167) | 2.6% |
SAP SuccessFactors partners | 2.0% / 3.8% | 0.0% (0 of 387) | 1.4% (5 of 350) | 16.5% (17 of 103) | 2.6% |
Table 2. Trade media’s share of citations by engine, prompts that don’t name vendors, 2 to 10 October 2026. Gemini’s SAP figure rests on 17 citations, mostly from two list-style sites. Google column: share of first-page results, 10 and 17 October 2026.
AI didn’t create the gap
Then I checked Google, twice, a week apart. Same questions, Google’s first page: trade media was between 0% and 10% of results both times, and vendor and company sites between 57% and 80%. Google’s own AI tools, AI Overviews and Gemini, treat trade media about the same as its search results, and in identity and endpoint management AI Overviews gave it more room than Google’s first page. ChatGPT, which isn’t built on Google, is harsher still.
Vendors have spent the best part of 30 years building their marketing around content, while publishers built theirs around traffic. The job of comparing suppliers went to vendors and review sites long ago. AI just makes it final.
So the gap was already there before AI arrived. What is new is that there’s no second page and no click left for trade media to close it.
Why being in the answer matters
For B2B media businesses, this goes straight to revenue. Advertising, sponsorship, lead and demand generation and newsletters are all priced on the evaluative audience: the readers comparing suppliers and deciding. A reader comparing suppliers is worth more to the vendors paying for your audience than one browsing the news.
When that comparing happens inside an AI answer that doesn’t name or cite you, you lose the reader and the proof you were ever part of the decision. And when investors and buyers come to value these businesses, they’ll price that exposure.
Vendors already write the buyer’s guide
Vendors’ own best-of, comparison and alternatives pages took between 12% and 39% of the citations that went to vendor and company sites: 39% in customer data platforms, 36% in HR and payroll, 33% in endpoint management, 22% in identity governance, 17% in financial services and 12% among SAP partners. In HR and payroll, three AI Overview answers linked one vendor’s name to a rival vendor’s “top 8” list. The rival’s own content was the source that introduced its competitor.
The answer isn’t more listicles
If you run a B2B media business, the lazy reading of this data is to write more rankings. Don’t. Thin lists win because nothing independent is there. It’s a numbers game publishers can’t win: every category has far more vendors than trade titles, each with a content team. And a thin list written to chase citations spends the one thing a publisher has that a vendor page doesn’t: independence. A ranking is only worth citing if readers trust it wasn’t bought.
The opening for independent B2B media is evidence and judgement a vendor can’t credibly offer: benchmarks with a published method, original data, tested comparisons, buyer’s guides the model can actually read. I won’t pretend the engines already reward depth. They don’t yet. But the slot is open.
The opportunity
This is the opportunity, and it’s a big one. AI goes looking for third-party judgement: in every category I measured, it cited review and analyst sites ahead of trade media, and when it names a reference, it names an analyst. When a trade title did publish a ranking, Google’s AI Overviews cited it again and again.
The published research points the same way. A University of Toronto study published in September 2025 found AI search leaning heavily on third-party sources over brands’ own content. A long-running study from a PR software company finds journalism behind roughly a quarter of AI citations. Both used consumer questions, the “rank the best smartphones” kind, and even there the Toronto team found brands’ own content gaining ground the closer a question got to a purchase. Enterprise buying questions are about as close to a purchase as questions get.
My own earlier work shows the same contrast. In a July 2026 pilot of UK business laptop questions, run with Otterly.AI, the four reported brand domains accounted for 11 citations between them, out of 205 cited URLs. The answers were assembled from tech media and YouTube. Where independent reviews exist, AI uses them.
The demand for independent, authoritative evaluation is already there. In enterprise B2B the supply isn’t, and no one is better placed to provide it than independent trade media.
Being cited doesn’t pay in clicks. Being named as the reference does.
Open, but not in control
44 of 46 trade titles whose robots.txt I could read leave the door open to the bots that fetch pages for live answers. For most, blocking isn’t the reason they’re missing. But open isn’t the same as in control. Most are letting AI fetch their pages with no say over whether they’re credited or paid, and as the ChatGPT numbers show, a page that gets fetched doesn’t necessarily get credited. Which bots get in, and on what terms, should now be a board decision.
If you market a vendor
Read the same data the other way round. When AI checks a shortlist, it checks you mostly on your own pages, including the best-of and comparison pages you wrote yourself. Those pages are now doing the job a trade review used to do. Make sure they’re accurate, current and specific enough to quote, and that every claim on them would survive a buyer checking it.
Does this apply to your sector?
I measured six enterprise technology and services categories, and I can’t assume the pattern holds in yours. In sectors with a deep bench of independent reviews and testing, it may run the other way. In others it may be worse. What you can’t afford is not knowing which way it works in your market. If you’re missing, the opening is the one described above. If you’re already the reference, the job is different: making sure you’re credited, and paid, for work the answer now does in place of your pages.
The question for B2B media boards
When a buyer in our sector asks AI for a shortlist, how often does it name or cite us?
Most boards can’t answer it, because nobody has measured it. If you run a B2B media business, email me at john@decoding-growth.com with one of your titles and the category it covers, and I’ll report back how often AI names or cites it on shortlist questions.
Frequently asked questions
A note on method
AI sources: tracked on a commercial AI search-tracking platform on ChatGPT, Gemini and Google AI Overviews, UK, one window of about a week per category between 22 September and 10 October 2026. Customer data platforms, HR and payroll and financial services include prompts that name vendors; identity governance, endpoint management and SAP partners use only prompts that don’t. Citations are raw counts for the latter three and estimated from retrievals and citation rates for the former three.
Classification: every cited domain was placed in one of six publisher types (trade and tech media; vendor and company sites; reviews, directories and analysts; community and social; government and institutional; reference and other), using the tracking platform’s domain classes plus Decoding Growth’s own rules. The same trade-media boundary was applied to the AI and Google sides. “Trade and tech media” includes business press and independent editorial sites, so the trade figures are an upper bound for trade titles alone. “Vendor and company sites” includes consultancies, systems integrators and law firms.
Mentions: answer text from the 27 prompts that don’t name vendors, across all six categories and three engines (783 answers), searched for about 120 trade and business title names. Generic-phrase matches were removed by hand.
Engine split: identity governance, endpoint management and SAP partners only, where citation data by engine was available.
Google: keyword versions of the same 27 prompts, Google UK desktop, top 10 organic results, pulled through DataForSEO on 10 and 17 October 2026. 78 of 81 queries returned results on 10 October (686 results) and 80 of 81 on 17 October (709). Every domain the rules didn’t recognise was checked by hand.
ChatGPT searches: the searches ChatGPT showed while answering, coded by hand, in two sectors.
Crawler access: robots.txt read for 52 trade titles, 46 readable. Declared policy only.
Limits
Six enterprise technology and services categories, chosen from active research projects, not a representative sample of B2B.
Three engines, one window per category. The Fraud Prevention Selection Index replicates the method on four engines plus Claude, with the prompt set fixed before the data.
Prompts were written by Decoding Growth and not tested with buyers.
Citations show what an answer leans on as evidence, not everything the model knows.
Google’s first page moves. Between a fifth and a half of each category’s results were the same in both pulls, so the Google figures by category are snapshots, not fixed shares.
About the author
John Webb is the founder of Decoding Growth. He measures how much of the B2B buyer’s decision now happens inside AI, what that does to the value of a business, and what to do about it. He has worked on every side of that decision. He ran B2B marketing at Samsung and Rackspace, was MD of Dennis Publishing and Autovia, and Executive Director at Spiceworks. Today he works with both sides: the media, events and information businesses that used to host the buyer’s comparison, and the B2B vendors that need to win it. The Selection Layer is his fortnightly newsletter and the model behind the work.
Connect on LinkedIn


