Backlinks didn't die. They got narrower.

Both classical SEO and AI citation are alive in 2026. Most sites still see 60–80% of organic traffic from Google's blue links, AI Overviews, and image search. Some categories (consumer commerce, professional advice, medical info, comparison content) are seeing AI Overviews eat 20–40% of click-through.

The SEO budget isn't either/or. It's allocation:

• HTML SEO still ships traffic today.
• AI citation ships brand and is the leading indicator of where traffic moves next.

You'd be wrong to dismiss either. But the leverage on each shifts continuously, and within both, what works has narrowed.

The classical model: backlinks = PageRank = ranking position. Volume mattered. Aggregate domain authority mattered. Ahrefs' DR / Moz's DA / Majestic's TF all rose with quantity. The link-building industry sold tier-3 PBN packages because the math rewarded volume.

What actually changed:

• Google's own ranking algorithms heavily devalued spammy / PBN / low-quality links over the last decade. The anti-spam updates kept narrowing what counted as a real link.
• AI grounding pipelines weight backlinks differently. They care about who links from where more than how many. Topical relevance dominates raw authority.
• A backlink from a respected sneaker enthusiast newsletter beats a backlink from NYT for a sneaker valuation site — even though NYT has 10× the domain authority. The model has learned that trust is topic-specific.

The volume-based playbook is dead. What replaced it: a small number of editorial backlinks from sources the model already trusts on your specific topic.

Two factors. Multiplicative, not additive.

• Vertical-trusted backlinks — editorial mentions from sources the model already trusts on your topic.
• Entity clarity — clean canonicals, complete schema, real freshness, disambiguated titles. (See One canonicalization rule.)

Each amplifies the other. A great entity page with no backlinks won't get cited — no one points to it; the model doesn't surface it. Five vertical-trusted backlinks pointing to a thin entity page won't get cited either — the AI lands but can't extract.

Score each backlink before pursuing it:

Backlink quality scorecard (per link):
  Vertical relevance:    0-3   (does the linker cover your topic?)
  Editorial nature:      0-2   (was this earned, not paid/PBN?)
  Domain trust on topic: 0-3   (does the model trust them HERE?)
  Anchor specificity:    0-2   (does the anchor describe your entity?)

Score 7+ → high-value, pursue
Score 4-6 → moderate, pursue if cheap
Score 0-3 → ignore or refuse

Most outreach campaigns chasing DR70+ generic links score 2–4 on this rubric. Most outreach campaigns chasing one editorial mention from a vertical newsletter score 8–10. Your backlink budget should fund the second.

Same shape, different layer. Classical keyword research: find high-volume head terms, target them, capture aggregate traffic. The volume curve was the whole game.

What actually changed:

• Head terms get eaten by AI Overviews. The click-through on "best electric SUV" cratered because the AI answers it directly above the SERP. Volume stayed the same; click-through halved.
• Long-tail, question-shaped queries are where citations happen. "best electric SUV under $40k for towing 5,000 lbs in upstate NY winters" still produces citations because no AI answer is comprehensive enough. The query is too specific to be pre-generated.
• The answerable claim replaces the ranking term. What you optimize for is being the source for a specific claim, not being top-3 for a generic term.

Keyword research today is closer to claim research. Three questions to ask of every term you used to chase:

1. What specific CLAIM do users phrase as a question?
   "Is X compatible with Y?" "What's the price of X in Z?"
   "Does X work for [narrow use case]?"

2. Who currently answers it well enough to be cited?
   What domains? What page formats? What schema types?

3. Can you answer more cleanly, with cleaner schema,
   on a more disambiguated entity page?

That's the new keyword research. It produces 30–50 answer-shaped opportunities per vertical, not 500 head terms.

What to actually do depends on how much authority you start with.

New domain (0–12 months in)

• Pick one narrow vertical you can plausibly dominate.
• Build the entity catalog — one canonical page per entity, full schema, the bifurcation pattern (see The second web).
• Earn 5–10 editorial mentions from vertical-trusted sources over 6–9 months.
• Distribute on platforms (Substack, vertical communities, app stores) for cold-start visibility.
• Ignore head terms. Target 30–50 specific question-shaped long-tail queries.
• Measure citation, not rank — see Rank vs citation vs grounding.

Mid-tier domain (1–5 years, some authority)

• Audit which existing pages are entity-worthy vs path-only. Most sites at this stage have 80% paths and 20% entities — concentrate effort on the 20%.
• Bifurcate the entity-worthy pages — add a markdown alternate via rel="alternate".
• Tighten topical concentration. Prune off-topic content if it dilutes your specialist signal. Models read the whole domain shape, not just the page.
• Earn 10–20 vertical-trusted backlinks per primary topic.
• Replace generic content with question-shaped answer pages.

Established domain (5+ years, strong authority)

• Defend canonicals. You have authority — don't fragment it across query-string variants. (Cardinality-aware canonical rules: see the canonicalization rule.)
• Audit citation share. You're probably losing citation to specialist sites on long-tail queries. Track this with a citation checker, not Ahrefs rank tracking.
• Add per-page markdown alternates on your top-cited entity pages first. The bifurcation pattern compounds.
• Refresh dateModifieddeliberately — it's a citation signal now, not just a sitemap field.
• Watch entity drift. AI grounding can shift to competitors silently if their evidence-fit improves and yours stagnates.

• Volume-based link building. Five vertical-trusted beats five hundred random — by an order of magnitude, not a percentage.
• Head-term obsession. "Best laptops 2026" is increasingly an AI Overview, not a click destination. You're ranking for impressions, not outcomes.
• Generalist content "to show authority." Specialist signal beats generalist breadth. The model concludes you're not a real specialist if you cover everything.
• Rank as primary KPI.Citation is the leading indicator. Rank still matters but it's lagging — the SEO who only tracks rank discovers problems three months too late.
• Keyword volume as primary input. Question-shape and answerable-claim are the sharper inputs now.
• Treating backlinks as inputs to PageRank. They're now inputs to candidate-set inclusion, training-corpus weighting, and grounding evaluation — three different layers, different mechanics each.

Both backlinks and keywords used to operate on volume curves. The strategy was aggregate: cover lots of terms, get lots of links, capture traffic at scale. Spreadsheets full of keywords; spreadsheets full of link prospects. The work looked like data entry.

Both now operate on concentrated curves. Strategy is specific: be the cited source for narrow claims, earn the trusted mentions in narrow verticals. The work looks like editorial — the three people in your vertical whose backlinks you actually want, the twenty answer-shaped questions your category genuinely needs answered better.

Stop counting volume. Start counting fit.

Five vertical-trusted backlinks plus a clean entity page beats five hundred random backlinks plus generic content. Twenty answerable-claim long-tails beat one head-term obsession. The work didn't disappear. It got more deliberate.

That's the new SEO. Same job. Sharper tools.

Instrumentation that pairs with this playbook:

• AI Citation Checker — measure citation share across ChatGPT / Claude / Perplexity for your prompt set.
• Entity Salience Audit — score each page's entity centrality so you know whether your evidence-fit is high or hostile.
• AI Bot Allowlist Checker — confirm the AI engines can actually fetch you. The precondition for any of the above.