The content AI actually cites (it's not what you think)

Welcome back.

Short one this week.

One chart, five links, and a founder making an unusual bet.

AI doesn't cite your opinions

Meltwater and LinkedIn Ads just published data on which content types AI assistants cite most. "Best X" listicles top the list at 54%. Side-by-side comparisons hit 50%. "How to choose" guides land at 33%.

Thought leadership? 8%. Dead last.

The LinkedIn post that gets 200 reactions does almost nothing for whether ChatGPT or Claude mentions your product when someone asks "what's the best tool for X."

The boring comparison page you keep putting off does almost everything.

Listicles and comparisons dominate as most-cited AI content formats.

There's a structural reason.

AI assistants answer buying questions, and buying questions want structured answers. Clear tables, stated prices, explicit verdicts. A comparison page is pre-formatted for extraction, but a hot take is not.

The move isn't to stop writing opinions, but to stop expecting them to do distribution work they were never going to do. Write the comparison pages for the machines and the opinions for the humans.

Which comparison page have you been avoiding? Reply and tell me. I'll probably nag you about it next week.

Four things worth your time

Base44 built its own model

The Wix-owned vibe-coding platform is rolling out its own AI model, hoping it will eventually outperform frontier models. The interesting part is the why: when everyone builds on the same three models, owning one is the only moat left.

Base44 lets anyone build apps using plain language.

Legora joins the legal AI billion club

Raised $550 million at a $5.55 billion valuation, with total funding now approaching $1 billion, a year after entering the US. Vertical AI keeps eating generalist lunch.

A VC-funded Tamagotchi for your habits

Momo Self Care raised $2.5 million for an app where completing personal tasks feeds a gamified digital pet. File next to HabitKit from two weeks ago: habit apps keep proving that simple mechanics beat clever ones.

Momo turns healthy habits into a virtual pet game.

GPT-5.6 Goes Public

OpenAI's GPT-5.6 (Sol, Terra, Luna) launched publicly today after a government-restricted preview. Sol leads on agentic coding and cybersecurity tasks; Terra offers GPT-5.5-level performance at half the cost.

ChatGPT Work launches, powered by the new GPT-5.6 model.

Teardown: the "build your own model" bet

Base44 is worth a closer look because it inverts the standard indie playbook.

The consensus move in 2026 is to be a thin, fast layer on top of frontier models and let OpenAI and Anthropic eat the compute costs. Base44 is spending Wix's money to go the other way.

AI Startups seek defensibility. Source: TechCrunch

The logic: if your product is "Claude with a nicer UI," your margin belongs to Anthropic and your differentiation belongs to nobody. A purpose-built model, even a worse one overall, can win on the narrow task it was trained for and cut inference costs at the same time.

For a solo founder this exact move is out of reach. The transferable version isn't the model, it's the question: what part of your product would still be yours if your AI provider tripled prices tomorrow?

If the answer is "the logo," that's your roadmap.