daily.dev's search traffic tripled in 90 days. Here is what we did, what actually moved the needle, and the lessons that contradict standard SEO advice.
Every week someone tells me SEO is dead. AI assistants answer questions directly, nobody clicks blue links anymore, stop investing.
Meanwhile, over the last 90 days, daily.dev's search traffic roughly tripled, to over 7,000 clicks a day. Our best month ever. And here's the part that makes this worth writing about: the people declaring SEO dead are not entirely wrong. Both things are true at once, and figuring out how they can both be true taught me more about search in three months than the previous three years did.
This is what we did, what actually moved the needle, and the lessons I didn't expect.
What we actually did
1. We merged our domains. This did more than two years of content.
For years, daily.dev the product lived on a subdomain (app.daily.dev) while daily.dev the website lived on the apex. Two properties, split authority, and hundreds of thousands of product pages, posts, discussions, and profiles that were invisible to the domain most people linked to.
We consolidated everything onto daily.dev. The effect was bigger than anything else on this list, and it wasn't close. Community pages started ranking. Post pages started showing up inside Google's AI features. The domain's link equity finally pointed at one place.
If you run a product on a subdomain and your marketing site on the apex, this is the single highest-leverage SEO decision you can revisit. It is also terrifying, weeks of redirect mapping, and you will find skeletons. Do it anyway.
2. We deleted and demoted more than we published.
The merge exposed years of accumulated debt: redirect chains, thousands of 404s that AI crawlers kept hitting, a tail of programmatic pages with zero citations and zero clicks, and a link-spam network that had attached itself to community features. We fixed the redirects, cleaned the junk, tightened what gets indexed, and disavowed the spam.
None of this is fun to ship or fun to tweet about. It showed up in the data within weeks.
3. We rebuilt content around evidence instead of instinct.
Every piece of content we publish now starts from data: what AI assistants actually cite, what people actually search, and where those two overlap (spoiler: barely, see lesson one). Refreshes must change substance. Swapping the year in a title and calling it updated does nothing, and we proved that to ourselves the hard way before believing it.
4. We started treating AI assistants as a first-class audience.
We maintain an llms.txt file, serve clean markdown versions of key pages, and explicitly welcome AI crawlers instead of blocking them. Whether you like it or not, a growing share of your "readers" are machines assembling answers for humans. We decided to be easy to read.
The lessons that surprised me
Lesson 1: what AI assistants cite and what Google ranks are two different worlds.
This is the finding I'd tattoo on every marketing dashboard. We looked at our top 50 pages by Google clicks and asked how many of them AI assistants cite when answering questions. Nine. The correlation between "gets Google traffic" and "gets cited by AI" on our content is statistically indistinguishable from zero.
Our most-cited pages by ChatGPT, Claude, and Perplexity get almost no Google clicks. Our biggest Google pages are nearly invisible to AI assistants. These are two separate games played on the same field, and you need a strategy for each. Today, daily.dev is one of the top three most-cited domains across the industry conversations we track, and most of the pages doing that work would look like failures in a classic SEO report.
Lesson 2: keyword tools are starting to measure robots, not people.
We found a keyword that every SEO tool swears gets around 20,000 searches a month in the US. We rank #2 for it. Search Console shows essentially zero clicks and barely any impressions. Where an honest 20K-volume term at position #2 should send hundreds of clicks a month, ours sent nothing.
The phrase follows a pattern ("[topic] news today") that humans rarely type but AI agents constantly do when they search the web on someone's behalf. Those agent queries flow into the clickstream data that keyword tools are built on, and the tools report them as human demand. My rule now: no surprising keyword volume enters a plan until it's reconciled against your own Search Console impressions. Divide impressions by the months in your window. That's your real ceiling. Everything above it is robots.
Lesson 3: a #1 ranking can be worth one visit a month.
We rank #1 for a commercial keyword with around 900 searches a month. Estimated traffic from that ranking: about one visit. Not one hundred. One. The AI Overview above the results answers the question, cites a few sources, and the classic results below it might as well not exist.
Ranking no longer implies traffic. On SERPs with AI Overviews, the return comes as impressions and citations, and if you judge those pages by clicks you will kill exactly the assets that are winning the new game.
Lesson 4: the demand curve itself is bending.
Across our entire topic space, search volume for the most recent month runs far below the 12-month average. One flagship term is at roughly a quarter of its yearly average. At the same time, our brand's citations inside AI answers have grown month after month, and AI-generated search features now show our pages about 100,000 times a day.
People are not asking fewer questions. They're asking them somewhere else. We can see the substitution happening inside our own data, which is exactly why "SEO is dead" and "our SEO tripled" can both be true: the classic channel is shrinking slowly while still being massively underexploited, and the AI channel is exploding while almost nobody measures it.
Lesson 5: the boring craft still pays, immediately.
For one high-intent query, our homepage ranked on page one with a click-through rate of 2.3 percent, roughly a third of what that position should earn. The reason was embarrassingly simple: our snippet never used the words people searched, so nothing bolded, so nobody clicked. We rewrote one meta description to speak the query's language. That's it. That's the fix.
While auditing, we also found broken English in the most trust-sensitive part of our browser extension listing, a database query that returned posts in effectively random order, and pages that rendered empty HTML to crawlers. None of this appears in any growth framework. All of it was quietly taxing us. Audits beat ideas.
What I'd tell another founder
Search isn't dying. It's splitting. There's the classic channel, blue links and clicks, which still delivered our best quarter ever the moment we fixed the fundamentals. And there's the AI channel, citations and answers, which runs on different content, different metrics, and different instincts, and which most companies aren't measuring at all.
The teams that win the next few years will run both playbooks and refuse to let one channel's scoreboard judge the other one's players.
And if you take a single thing from this post: before you write one more piece of content, merge your subdomains, fix your redirects, and read your own Search Console like it owes you money. The least glamorous work we did was the most valuable. It usually is.
