Emily Kramer spent a while debating whether the website is still the most important marketing asset, given how much research now happens inside a chat window. She landed on yes, but for different reasons than before.
We should say plainly that we are in this one. Casey supplied most of the examples and DoWhatWorks data sits behind several of the claims. Read it with that in mind, then read it anyway, because the framing is better than ours.
The argument
Same asset, different reasons
Less traffic reaches your site because people research in a model first, and the traffic that does arrive is harder to convert because opinions are half formed before the click.
Your website is a major source of what feeds LLMs, so what is on your site shapes what those models say about you.
The second argument is differentiation. When anyone can vibe code a decent looking site in an afternoon, the site is your chance to not look like everyone else. A Framer report found 83 percent of marketing teams manage multiple sites, which is how the chaos actually happens.
Bucket one
You are writing for two audiences now
This is not the same as optimising for search. With search you controlled the title and the meta description. A model writes a different answer every time, so you have to guide it rather than instruct it.
Prompt presets Wispr Flow ends the homepage with still not sure, let ChatGPT, Claude or Perplexity do the thinking for you, and three buttons. Confident enough to let machines do the talking. Glean and Galileo run versions of the same thing.
Machine readable files Supabase links humans.txt, lawyers.txt and security.txt from the footer, giving crawlers a clean endpoint instead of making them scrape an About page.
Content that suits both Kick lists tax deadlines by entity type with add to calendar buttons. Useful to a human, trivially citable by a model. Plain puts the changelog in the hero, which signals shipping speed and keeps the page fresh.
On comparison pages, GC.AI compares itself with ChatGPT and Claude in a table on the homepage. Framer has a dedicated Claude comparison page that was ranking inside ChatGPT within a month. PostHog goes furthest, with a side by side that admits where a competitor suits you better and a page titled you will hate PostHog if.
Bucket two
Make the proof verifiable
Most sites still do social proof the way they did five years ago. A logo bar, a quote carousel, a case study page buried in the nav.
Clay was early to fixing it. Hover a logo and you get a quote from a named person with a role and a photo, labelled case study or hackathon so you know what the click gives you. HeyReach goes further, showing performance metrics and the other tools that customer runs alongside it.
Beyond the logo bar, Linear organises customers in a filterable table by industry with a read story column, CentralHQ links each testimonial to its original post on X, LinkedIn or G2, and Clay and Sierra both give security certifications real homepage space.
The more vague and high level the social proof, the worse it performs.
Bucket three
Make starting easier than deciding
Lovable put a model style input bar in the hero and arguably started the trend. Framer runs dual CTAs, start for free and start with AI. Mutiny rotates prompt ideas instead of a typeable box.
Templates moved to the homepage too. Replit shows presets like B2B project management app below the input. Notion pairs copy and paste prompts with a try it button and a walkthrough.
The migration examples are the sharpest. Pylon segments its social proof bar by migrated off Zendesk, migrated off Intercom, and others, which is positioning, proof and reassurance in a single element.