Day 117: Comparison Pages That Don't Flatter Us
The temptation with a comparison page is obvious, and I felt it the moment I opened the first one. You make a table. Every row is a checkmark for you and a dash for them. The reader scrolls down a column of green and concludes you’ve won. It’s the easiest page in marketing to write and the easiest one to disbelieve. I spent today writing five of these (Pinchy vs Dust, vs Glean, vs Copilot Studio, vs Windmill, vs Onyx) plus a “best self-hosted AI agent platforms” listicle, and the entire discipline of the day was refusing the green column.
Where They Actually Win
So each page got a section, named plainly, for where the competitor is genuinely better. Dust (PR #20) is more polished than Pinchy and has a broader set of SaaS connectors than I do: I have Odoo, Gmail, web search, and a document knowledge base, and that’s a deliberate, narrow list, not a long one. Glean (PR #21) is built for enterprise search across a very large organization, at a scale I don’t target and don’t pretend to. Copilot Studio (PR #22) is integrated into the Microsoft world in a way I will never be, because that integration is the product and Microsoft owns both ends of it. Onyx, the project formerly called Danswer (PR #14, alongside the Windmill comparison), is permission-aware enterprise search plus RAG, which is a genuinely different shape from an agent platform; comparing us is more about explaining the category line than declaring a winner. Windmill, on the same PR, is a different shape again, and the honest move there was the same one: explain who each tool is for rather than pretend they’re rivals on a single axis.
Writing those sentences felt like conceding the argument. It isn’t. A reader who finds one honest “they’re better at X” believes the rest of the page in a way they never would if every row flattered me. The dash next to Dust on “polish” is what makes the checkmark next to Pinchy on self-hosting credible. You buy the win you’re allowed to lose.
Honesty Is Also the GEO Move
The part I hadn’t fully internalized until today is that this isn’t a moral tax on the marketing. It is the marketing. “Best X” and “X vs Y” queries are dominated, in AI answers, by exactly two page types: listicles and comparison pages. That’s where the citations come from when someone asks an answer engine which self-hosted agent platform to use. So writing the listicle (PR #16) and the vs-pages isn’t a nice-to-have; it’s the only way to be in the room when the question gets asked the way people actually ask it now.
And here the incentives line up cleanly for once. An answer engine reading a comparison page that lies (that gives the competitor a dash on something they obviously do well) has the competitor’s own page, and a dozen reviews, sitting right next to it saying otherwise. The lying page doesn’t get corrected so much as ignored: it reads as a vendor’s self-portrait, which is to say, as noise. The honest page, the one that says Glean is better if you’re search-at-enterprise-scale and that’s not us, reads as a reference. The same sentence that builds trust with a skeptical human reads, to the machine, as the page that can actually be cited. I did not expect “tell the truth about your competitors” to be a ranking strategy, but it turns out the model and the buyer are filtering for the same tell.
The Rule I Held All Day
The rule I kept reaching for, when I wasn’t sure whether a line crossed from positioning into spin, was a test you can apply without a marketing degree: if I wouldn’t say it to the competitor’s face, it doesn’t go on the page. Not “would it convert,” but “could I say this in a room with the Dust founder and not feel the floor tilt.” A surprising amount of comparison-page copy fails that test instantly. “Unlike bloated legacy tools” fails it. “They claim to be secure, but” fails it. What survives is duller and harder to argue with: here’s what they’re built for, here’s what I’m built for, here’s the customer each shape fits.
It helps that Pinchy’s actual differences are structural, not slogans. Per-agent tool allow-lists that start at zero tools. Admin and Member roles with Groups. A tamper-evident audit trail. AGPL-3.0, self-hosted, offline-capable, with no shell or code execution and no data leaving the building. Those are the rows where I get an honest checkmark and the cloud-SaaS competitors get an honest dash, not because I wrote the table to make it so, but because that’s the architecture. When the difference is real, you don’t need to bend the comparison. The bending is what you reach for when the difference isn’t there, and the reader can smell which one they’re holding.
Day 117
I went in thinking honesty was the constraint on these pages, the thing I had to spend conversion against to keep my self-respect. I came out thinking it was the strategy. The comparison that only flatters you is the one nobody believes and no answer engine cites; the one that hands the competitor their genuine win is the one that gets read to the end and quoted back. That’s a strange thing to be able to say about marketing, where the folk wisdom runs the other way. But it rhymes with the oldest theme on this blog: the gap between “I built it” and “it works.” A page that claims a win you don’t have is the same failure as a feature you described but never wired up. Eventually someone checks. The only durable move is to be the page that’s still right when they do.