For fifteen years the SEO advice for a new site was the same: write something, then go get links to it. Guest posts, directory submissions, "skyscraper" outreach, the lot. If you weren't building links, you weren't doing SEO.
In 2026 that advice is out of date. Links didn't stop counting. What changed is what they predict, and what founders are told to chase — domain authority — is the part that stopped predicting almost anything.
We didn't come to this from a hot take. We came to it by measuring our own sites and being wrong in public first.
Key takeaways
- Backlinks still matter, but at the page level, not the domain level. How well-linked the specific pages ranking for a query are still tells you something. How big the domains are told us less than nothing.
- Topic selection beats link building for a young site. Most of the search demand that ended up reaching our site sat on queries that keyword tools didn't even list.
- Clusters compound. Publishing a tight group of related articles raised impressions per page roughly eightfold on the same site — no link campaign involved.
- The competition you can actually beat is the one link-based metrics can't see: the page type already ranking (a tutorial vs. a tool homepage) decides more than its authority score.
- Be suspicious of any single number — including ours. One site's data is a hypothesis, not a law.
What "backlinks matter" used to mean
Early Google leaned heavily on links because they were the best proxy it had for "people vouch for this page". So the whole industry built tooling around counting them: domain rating, domain authority, keyword difficulty scores that are, underneath, a count of the links pointing at whatever currently ranks.
That created a simple mental model for founders: big sites win, so become a big site, and links are how you become big.
The trouble is that the model got baked into tools long after the search results it described stopped looking like that.
What we actually measured
Grove plans what to write for every customer site, so we have a direct stake in knowing which numbers predict rankings. We graded the standard metrics against real Search Console history on a design-tool site we operate: 171 non-branded search queries, checking which of the metrics would have told us, in advance, whether the site could reach the top 20.
The result surprised us enough that we rolled back a feature the same day we shipped it:
| Signal | Did it predict reaching the top 20? |
|---|---|
| Keyword difficulty (links pointing at the pages that rank) | Somewhat better than a coin flip |
| Average domain authority of the sites ranking | Worse than a coin flip — it pointed the wrong way |
| Our own blend of the two | Worse than difficulty alone |
Read that middle row again. The queries where the ranking sites had huge domains were often the ones we ranked for. The reason was obvious once we looked at the pages instead of the scores:
- "How to do X in Photoshop" searches are dominated by enormous domains — Adobe's docs, YouTube. But a specific, well-written tutorial still gets in at positions 7–10, because it answers a narrower question better. Big domains, beatable page.
- "Auto background remover" searches were full of smaller brands with lower domain scores. But every result was a dedicated tool homepage with plenty of links pointing at that exact page. We sat at position 40–68 there and never moved. Small domains, unbeatable page.
Domain authority got both of those backwards. Page-level link counts got them roughly right. That is the whole story of backlinks in 2026 in one example: links still describe how entrenched a specific result is. They no longer describe how strong a site is in a way that decides anything for you.
Why the shift happened
A few things changed at once, and none of them requires believing a rumour about the algorithm.
Search results became a mix of page types. A single query can return a forum thread, a video, a tool, a vendor doc and an AI-generated summary. A link count can't tell you which of those types Google wants for that query — and the type is usually the real barrier.
Answers moved above the results. AI summaries at the top of the page, and assistants like ChatGPT answering directly, reward the page that states the answer clearly and specifically. Being citable is a different skill from being linked to. (We wrote up what that takes in How to Get Cited by ChatGPT (And Why It Matters More Than Backlinks).)
Most real demand is in the long tail, where links barely exist. On the site we measured, 72% of non-branded impressions came from queries the keyword database didn't know about at all. Nobody is building links to pages targeting those queries, because nobody's tool says they exist. That's where a young site can win.
What to do instead of chasing links
If you have a small site and limited time, this is where the effort pays back faster.
1. Pick topics by the page type you can beat, not the authority score
Before writing, look at what already ranks. If it's ten tool homepages, a blog post won't break in no matter how many links you build. If it's a mix of tutorials and forum threads, a sharper answer can. Our six pre-writing checks start with exactly this.
2. Use your own Search Console as the keyword tool
The queries Google already shows your site for — even at position 30 — are the best research you have. They're real demand, they're rarely in keyword databases, and Google has already decided you're relevant. See Keyword Research for Founders: How to Rank Without Touching Ahrefs.
3. Build clusters, not one-off posts
On the same site, as the number of articles in one topic cluster grew from one to ten, impressions per page rose from about 7 to about 57. More pages on the same subject didn't split the traffic; each page did better because the site became the obvious source for that subject. That's topical authority doing what links used to be credited for. The mechanics are in How to Build a Hub and Spoke Content Strategy.
4. Earn the links that come for free
None of this means refusing links. A genuinely useful page — a benchmark, a template, an honest comparison — gets linked without outreach, and those links help that page. What we'd stop doing is spending weeks on outreach campaigns for a new site whose real problem is that it's targeting the wrong queries.
The honest caveat
This is one site's data. Search Console only shows queries you already appear for, so it's biased toward winners, and a few hundred queries is evidence, not proof. We'd be doing exactly what we're criticising — trusting one number — if we told you backlinks are dead.
They're not dead. They've been demoted: from the strategy to one signal among several, and a weaker one than the choice of what to write in the first place.
FAQ
Do backlinks still matter for SEO in 2026? Yes, but mainly at the page level. How well-linked the specific pages ranking for a query are still indicates how hard that query is. Site-wide authority scores predicted rankings poorly in our data.
Should a new site do link building? Not first. Pick queries where the ranking pages are beatable, publish a focused cluster, and let useful pages attract links naturally. Outreach makes more sense once you know which pages are close to page one.
Is domain authority useless? As a predictor of whether you can rank for a query, it performed worse than chance in our measurement. It's still a reasonable rough description of how established a site is — just not a decision-making number.
What matters more than backlinks now? Choosing the right queries (especially long-tail ones keyword tools miss), matching the page type Google already rewards for that query, and covering a topic in depth across several related articles.
Does this apply to AI search too? Largely, yes. AI answers tend to cite pages that state a specific answer clearly. That rewards clarity and depth on a narrow topic more than raw link counts.
