I read filings for a living, so I found this in a 10-K rather than in a marketing deck. On 12 February 2026 Gartner filed its annual report. Inside it, a line recording the sale of its Digital Markets division for approximately $110.0 million, closed on 5 February. Digital Markets is Capterra, GetApp and Software Advice. The buyer was G2, announced 29 January. Combined, those properties carry around 6 million verified reviews across more than 2,000 categories and claim more than 200 million software buyers a year.
A hundred and ten million dollars is a rounding error on Gartner’s balance sheet. It is also the price of the property that now decides which three vendors get named when a buyer asks an AI engine what to buy. I have never once seen that property appear on a marketing plan as an asset with an owner.
The thesis
Your content budget is pointed at the property that only pays after a buyer already knows your name. The property that produces buyers who do not know your name is owned by somebody else, and this year it started consolidating. This is not an authority problem to be solved with more publishing. It is an ownership problem.
Two populations of prompt, and almost nobody separates them
Here is the thing that makes the AEO debate go in circles. Two credible studies published within a year of each other say opposite things, and both are right.
Profound classified 11.84 billion citations between 16 April and 16 July 2026, across eight engines: ChatGPT, Claude, Google AI Mode, Google AI Overviews, Gemini, Grok, Microsoft Copilot and Perplexity. It mapped 3.02 million domains covering 98.3 percent of citation volume. The headline: about 57 percent of citations go to brand sites. In 24 of 29 industries, the brand’s own domain is the largest bucket for the median company. Cybersecurity runs at 74 percent brand. Marketing runs at 72 percent.
AirOps looked at 21,311 brand mentions across more than 500 commercial-intent queries on GPT-5, Claude Sonnet 4.5 and Perplexity Sonar. Queries of the form “best HR software” and “top CRM platforms”. The headline: 85 percent of brand mentions came from external domains. Only 13.2 percent came from the brand’s own site. A brand was 6.5 times more likely to be mentioned through somebody else’s page than its own.
Fifty-seven percent versus thirteen percent. The gap is not a methodology dispute. It is the prompt.
Profound measures the prompt sets its customers track, and companies that buy a visibility tool track their own name. When your brand is in the question, the engine goes to your documentation, your pricing page, your changelog, and it should. AirOps measured category questions with no brand name in them. When your name is absent from the question, your pages are not candidates for the answer. They are candidates for the follow-up.
So the owned-media spend is not wasted. It is conditional. It pays on the prompts that already contain the thing you were trying to buy.
The unbranded prompt is where the new revenue is
G2 surveyed 1,076 B2B software buyers and decision-makers in March 2026 and interviewed 39 marketers alongside it. 51 percent now start software research with an AI chatbot more often than with Google, up from 29 percent in April 2025. Within seven months, the share calling chatbot research more productive than search went from 36 percent to 53 percent.
Two numbers from that survey matter more than the adoption curve. 69 percent chose a different vendor than the one they planned to choose, on chatbot guidance. 33 percent bought from a vendor they were not previously familiar with.
Read that second figure as a prompt type. A buyer who was not familiar with you did not type your name. One in three purchases now begins in a question your own pages cannot be cited in. That is the population your content budget does not reach, and it is the only population that grows your customer list rather than confirming it.
The Content Marketing Institute surveyed 1,015 B2B marketers between 24 June and 14 August 2025 on where 2026 investment goes. Owned media came in at 32 percent of top-three selections. Social and earned media came in at 20 percent. The budget is weighted 1.6 to 1 towards the property that works on the branded half.
Three readings of the same data
Software has the thinnest earned base in the market. Profound puts the median earned share for SaaS and software at 11.4 percent. Set that against government and nonprofit at 15.9 percent brand share, which is to say almost everything cited about them sits somewhere else. Software companies publish the most and are the most dependent on their own domain, in the category where a third of buyers just bought from somebody they had not heard of. Inside a single industry, the spread between the 75th and 25th percentile company on earned share reaches 43 percentage points. That is not a market average problem. Some companies in your category have already made this trade.
The shape of the third-party page is specific, and it is not a blog post. Nearly 90 percent of those third-party mentions sat in listicles, comparisons or reviews. About 80 percent of the brands cited in them appeared as one of the first three companies discussed. Position inside somebody else’s page is the unit. Not the page you wrote. Not the word count.
The engines do not agree, so neither should your plan. ChatGPT runs 47 percent brand and 30 percent earned, the highest earned share of the set. Gemini runs 69 percent brand. Google AI Overviews sits at 17.0 percent earned and 15.3 percent social, citing social sources at roughly 1.3 times ChatGPT’s rate and 4 times Microsoft Copilot’s, where social is about 1 citation in 29. On first-party share, GPT-5 ran 4 to 11 percent while Claude and Perplexity ran 13 to 21 percent. And 68 percent of brands appeared on only one platform at all. A single content strategy, measured as a single score, averages four different games into a number that describes none of them.
Then the ownership fact lands on top. 45 percent of those 1,076 buyers said a citation from a software review site is the most confidence-inspiring signal in an AI answer, the highest of any source. Review sites are also the only source besides the chatbot itself whose influence rises deeper in the funnel, from 40 percent at discovery to 47 percent at retention. That is the signal buyers trust most. As of 5 February it has one fewer landlord than it had in January.
Where this breaks
Four honest limits, and the first two run against me.
Profound’s 57 percent is real and it is the largest sample here by three orders of magnitude. If your category’s demand genuinely arrives through branded prompts, your owned-content budget is correctly placed and this post is an expensive distraction. Cybersecurity at 74 percent brand looks exactly like that category.
AirOps sells the platform its report recommends, the report gives no data collection window, and it ran on GPT-5, Claude Sonnet 4.5 and Perplexity Sonar. It did not touch Google’s surfaces, which carry far more query volume than all of those combined. The 85 percent is directional, not a constant.
G2 asked buyers who use G2 whether they trust review sites, and G2 bought three more review sites a month before fielding it. I am using the number anyway, because the direction matches citation data collected by people with no review marketplace to sell. Weigh it accordingly.
And the limit I cannot close. Everything above shows where the citation slot sits. None of it shows that being named in a comparison page causes a purchase. Citation share is an availability measure. Moving budget on it is a bet on a correlation, and anyone who tells you the attribution is solved is selling something.
What I would do on Monday
Split your tracked prompt list in two: prompts that contain your brand name, prompts that do not. Report the halves separately and never again as one score. If your tool will not do this, export and do it in a spreadsheet.
For the unbranded half only, pull every cited URL from the last 30 days and add one column: owned, or not owned. That column is the whole diagnosis and it takes an afternoon.
Take the top 20 not-owned URLs and write the owner and the access mechanism next to each. Some are a review profile you control. Some are an editor who takes a pitch. Some are a forum thread nobody controls and you need to stop pretending otherwise.
Treat your review profiles as a product surface, not a vanity page. Category placement, review recency, response coverage, feature tagging. You are being read by a retrieval system that counts position, and 80 percent of cited brands are in the first three named.
Invert one slice of the 1.6 to 1 split for a single quarter, and judge it on unbranded citation share rather than sessions. If it does not move in 90 days, you have an answer worth having.
Close
The filing was public on 12 February. Anybody could read it. A $110.0 million transaction changed who owns the pages that introduce vendors to buyers who have never heard of them, and the teams whose pipeline depends on those pages were not in the room and did not notice.
You cannot buy the property now. You can at least find out whose it is.
Most AEO programmes I get asked to review are measuring one number that averages the half they can influence with the half they cannot. If you want your prompt set split into branded and unbranded, and the cited URLs marked owned or not owned, my calendar is here.
Splitting prompts into branded and unbranded turns a sterile “which study is right” debate into a measurement design question. The two numbers stop contradicting each other once you see they answer different questions. Most AEO debates never ask which prompt population a number describes.