Founder Decision Ledger #1: Reddit Wins 7 of 13 "Is It Worth It" AI Answers. My Site Wins One.

I killed a budget line this week. The page it was funding is still up, it still converts, and I am not touching it. I just stopped paying to make machines cite it, because I finally had the data to see that they never will.

I coined Machine Relations because the old vocabulary for visibility stopped describing what I was watching. Then I built the instrument that measures it, and I run the company that sells against those numbers. That means the instrument gets to tell me I am wrong. This week it did.

Here is what it told me.

Concede the part that is true

Your website earns citations. That is real, and anyone who tells you your own domain does not matter has not looked.

In the current release of the Machine Relations Index, vendor-owned domains take 6.9% to 13.8% of the top-ten cited slots depending on which question a buyer asked. In head-to-head comparison answers they hit their peak. Companies like IBM, Databricks and Palo Alto Networks hold comparison slots in their own categories. Your specification pages, your integration docs, your pricing table: machines read them and cite them.

So the website works.

For one question.

Six questions, six different markets

Here is the layer most founders never look at. "AI visibility" is not one thing you win or lose. The index splits buyer demand into six question shapes, and they behave like six separate markets with six different winners.

The six shapes are best tools in a category, how buyers choose, is it worth it, problem-first research, ranked top lists, and head-to-head comparisons. Same buyers. Same category. Same 125 days. Different citation winners in each.

I pulled the top-cited domain in every published segment and counted which class of source held it.

Question shape	Community platform	Vendor's own site	Editorial	Other classes
Is it worth it	7 of 13	1 of 13	0 of 13	5 of 13
Problem-first research	7 of 13	0 of 13	0 of 13	6 of 13
Best tools	3 of 13	2 of 13	4 of 13	4 of 13
How buyers choose	2 of 13	1 of 13	2 of 13	8 of 13
Head-to-head comparison	1 of 13	1 of 13	0 of 13	11 of 13
Ranked top lists	1 of 13	1 of 13	3 of 13	8 of 13

Read the bottom rows and the top row together. Your own domain wins the top slot once, in every shape, all the way down the table. It never wins twice. The shape where vendor domains take the largest share of the full top ten is the comparison question at 13.8%, and even there they take the single best slot exactly once.

Now read the top row again.

The question that decides the money

"Is it worth it" is the last question before a buyer spends. It is the one I care about more than any other. It is also the one I have the least ability to answer.

Across the 13 categories that publish a rate for that shape, reddit.com holds the top cited slot in 7. No vendor-owned domain holds it except one. Here is every winner.

Category	Top cited domain	Class	Citation rate
AI infrastructure	reddit.com	Community	32.1%
Consumer finance	reddit.com	Community	31.8%
Consumer products	reddit.com	Community	27.9%
AI visibility and GEO	reddit.com	Community	27.7%
Emergent prosumer	reddit.com	Community	25.0%
Education and training	reddit.com	Community	23.0%
Family software	reddit.com	Community	21.0%
Consumer health	uclahealth.org	Unclassified	37.7%
Deep-tech hardware	vlsishuttle.com	Unclassified	35.2%
Cybersecurity	paloaltonetworks.com	Vendor-owned	28.2%
Enterprise software	erpresearch.com	Market database	24.3%
Fintech	paymentsandrisk.com	Unclassified	23.2%
AI security and privacy	witness.ai	Unclassified	21.5%

Seven of thirteen. One forum.

In seven of those thirteen categories, not a single vendor-owned domain appears anywhere in the top ten for that question. Not the leader. Not tenth. Across all 13 segments there are 130 top-ten slots and vendor-owned domains hold 12 of them.

I have been funding a page whose whole job is to answer "is this worth the money." Machines answer that question 1,522 observed times in this window and they go to a forum to do it.

What the buyers said, before the machines agreed

The uncomfortable part is that this is not new information. It is old information I had been reading as a marketing problem instead of a structural one.

TrustRadius found in its 2026 Buying Disconnect report that 63% of buyers used AI during their purchase journey, 94% of those fact-check what it tells them, and 74% still use reviews to decide. HG Insights summarized the same study as trust more, verify everything.

SurveyMonkey and Reddit published research in March 2026 that puts peer recommendations at 73% trust, ahead of vendor websites at 55%, search engines at 54%, review sites at 46% and AI chatbots at 39%. Reddit co-sponsored that study, so weight it accordingly. Weight it anyway, because it agrees with a measurement Reddit had nothing to do with.

G2's 2026 buyer behavior research found review sites at 38% overtaking AI chatbots at 37% as the source shaping a shortlist, with finance involvement in software decisions jumping from 31% to 46% in a single year.

Buyers said they trust peers over vendors. The machines that read the web on their behalf now cite peers over vendors on exactly the question where that trust matters most. Two independent methods, one answer.

The exception is the whole lesson

One category breaks the pattern, and it is the most useful thing in this release.

In cybersecurity, five of the ten most-cited domains for "is it worth it" are vendor-owned, and a vendor holds the top slot outright.

Domain	Class	Citation rate
paloaltonetworks.com	Vendor-owned	28.2%
cyware.com	Unclassified	25.2%
reddit.com	Community	19.1%
swimlane.com	Unclassified	15.3%
sentinelone.com	Vendor-owned	14.5%
breachsense.com	Market database	13.0%
microsoft.com	Vendor-owned	13.0%
rapid7.com	Vendor-owned	13.0%
techtarget.com	Editorial	13.0%
ibm.com	Vendor-owned	11.5%

Look at what those five vendors are not. They are not uniformly enormous. Microsoft sits at rank 12 of 22,179 cited domains with an A confidence grade and citations in 13 categories. Rapid7 sits at rank 467 with a C grade and citations in 2. Rapid7 is cited in 31 of 15,782 observed runs across the whole index, which is a rounding error, and it holds a top-ten slot on the question that decides a cybersecurity purchase.

Overall authority is not what put them there. Concentration is. Every one of those five is cited inside cybersecurity and its adjacent security category, and the two smallest of them publish in almost nothing else.

That is the move, and it is not the move I would have guessed.

The decision, and the four parts of it

One. The worth page comes out of the visibility budget and stays in the sales budget. It closes deals for humans who already found us. It is not a citation asset in a category where no vendor domain holds a top-ten slot on that question. I was paying twice for one job.

Two. I read my own category before I copy anyone. Seven of thirteen categories have zero vendor presence in the worth answer. One has five. A founder in cybersecurity and a founder in family software are playing two different games, and the playbook that made one of them look smart will make the other look busy. Pull your category's segment page, for example cybersecurity worth answers, and read the top ten before you spend anything.

Three. Money moves to the shape my domain can actually hold. That is the comparison question, where vendor-owned domains take 13.8% of top-ten slots, roughly double their share in the best-tools shape. I would rather own the question I can win outright than rent third place in the one I cannot.

Four. For the worth question, I stop publishing and start being cited. Not posting in forums under a company account, which is the fastest way to get a thread removed and deserve it. The input is customers who say specific numbers in public. That is a product and support problem before it is a marketing one, and it is slower than everything else on this list.

The ledger entry

Every decision I make from this data gets recorded the same way, including the condition that kills it. If I only publish the decisions that aged well, this is marketing, not a ledger.

Field	Entry
Decision	Remove the worth/ROI page from the AI visibility budget; reallocate to the comparison shape
Date	2026-09-18
Evidence	MRI release mri_score_v2.0, generated 2026-09-18, window 2026-05-10 to 2026-09-18
Scope	13 published "is it worth it" segments, 1,522 observed runs
Reversal condition	Any release where vendor-owned domains hold the top slot in 4 or more of the 13 worth segments, or where our own category shows a vendor domain in the top ten
Review	Next quarterly release read

If that reversal condition fires, I was wrong and I will say so here.

What this measurement is and is not

The index observes six engines answering the market's buyer questions: ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews and Perplexity. This release covers 15,782 answer runs, 124,397 citation events and 22,179 cited domains over 125 days. A segment publishes a rate only after 10 observed runs across 7 distinct dates, so thin signal is never shown as settled. The machine-readable release carries every number above.

Two limits matter. The index runs the market's questions, not questions about any one brand, so it cannot tell me whether my pages are performing. And citation is correlation. Nothing here proves that a page caused a citation rate, and nobody should publish that it does.

One more thing worth knowing about the source that wins this question. Reddit was cited by four of the six engines in this window, not all six. Google's own crawler documentation describes Google-Extended as the control governing Gemini grounding, separate from Search inclusion, and Google's AI features guidance describes those surfaces drawing on the ordinary index. Perplexity documents query-time search over live sources. Each engine assembles its own answer from its own retrieval. The domain that dominates the worth question in four engines is not in the other two.

The binary

You are funding one of two things right now.

Either you are paying to answer a question the machines already decided not to ask you, or you picked the shape you can hold and put everything into owning it.

There is no third option where you win all six. The data says nobody does.

FAQ

What is a question shape? A pattern the Machine Relations Index groups buyer queries into. There are six: best tools in a category, how buyers choose, is it worth it, problem-first research, ranked top lists, and head-to-head comparisons. Each one is measured separately per category.

Does this mean my website does not matter for AI citations? No. Vendor-owned domains hold 13.8% of top-ten slots in comparison answers, their strongest shape. It means your website matters for some questions and not others, and the worth question is the one where it matters least.

Why does Reddit win the worth question? The index reports what engines cited, not why. The pattern lines up with what buyers report about trust, where peer recommendations score 73% against vendor websites at 55%. That is consistent, not causal.

Is cybersecurity really different, or is that noise? Five of ten top-cited domains in that segment are vendor-owned against a median of zero to two elsewhere, over 131 observed runs and 7 distinct run dates. It clears the same evidence floor as every other published segment.

How do I check my own category? Open the Machine Relations Index, find your category, and open the "is it worth it" segment. Read the top ten and count how many are domains you could ever own.

Can I reproduce these numbers? Yes. The release is public. Filter domains[].mri_score_v2.strata[] to status: "published" and question_shape: "is_x_worth", rank each category by citation_rate, and join source_role from the domain record.