AI Overviews compress the search results page. ChatGPT, Claude, and Perplexity have become real search interfaces for a meaningful slice of users. The SEO playbook has genuinely changed — not by erasing the old rules but by adding a new layer on top of them. Here is what survives, what’s new, and what to actually do.
What changed: the SERP got compressed
A 2022 Google results page for “best CRM for small business” showed ten blue links plus maybe an ad block. A 2026 results page for the same query shows: an AI Overview synthesis at the top (citing 3-5 sources), then a People Also Ask block, then maybe four or five blue links, then a video carousel. The classic ten-link list is gone.
What this means for click-through rates: the top three sources cited in the AI Overview capture most of the residual click traffic. Positions 5-10 in the classic ranking now receive a small fraction of what they did three years ago. The skew toward the top is sharper than ever.
What this means for ranking strategy: being in positions 1-3 organically is now table stakes for AI Overview consideration. Being beyond position 5 is functionally invisible on most informational searches.
What didn’t change: the fundamentals still rule
The work that gets you into AI Overviews is essentially the same work that got you into position 1 in the classic SERP:
- Substantive content depth.Pages that thinly cover a topic don’t get cited. Pages that answer a specific question definitively, with supporting evidence and examples, do.
- Topical authority. Single articles in a domain that has no other coverage of the topic underperform domains with multiple linked articles forming a cluster.
- Schema markup. Article, FAQPage, HowTo, Product, Organization, LocalBusiness schema all matter; they help both classic crawlers and AI training pipelines understand what the page is about.
- Backlinks from credible sources. Still a primary trust signal. Editorial mentions in trade press and adjacent industry blogs.
- Fresh updates.Pages that haven’t been touched in 18 months deprioritize. Quarterly updates with new evidence keep authority current.
What’s changed isn’t what works — it’s the consequence of not doing it. In 2020 a thin page might rank position 6 and pick up some traffic. In 2026 that same page ranks position 8, gets bypassed by the AI Overview, and receives essentially zero clicks.
What’s new: Generative Engine Optimization (GEO)
Beyond Google’s AI Overviews, a separate question matters: when users ask ChatGPT, Claude, or Perplexity about your industry, do those answers cite or recommend your site?
This is what GEO is. Some elements overlap with classic SEO; others are distinct:
- Be the source of an idea.LLMs cite pages that originate frameworks, methodologies, and specific opinions. Aggregator pages summarizing others’ ideas get cited less than the original.
- Use clear headings and structured lists. LLMs parse heading hierarchy aggressively to extract context. A page with H2/H3 questions and direct answers below them is highly extractable; a page with no structure is less so.
- Write definitively, not equivocally.“The answer is X because Y” gets cited. “It depends on many factors; here are some considerations” gets summarized away.
- Build entity associations. When your brand name appears alongside topic keywords across many credible pages on the web, LLMs learn the association. Press mentions, industry quotes, expert interviews all contribute.
- Track citations.Tools like Profound, AthenaHQ, and SE Ranking Visibility now offer LLM-citation tracking. If you can’t measure your citations, you can’t optimize for them.
The practical changes to your content strategy
1. Pillar + cluster structure becomes more important, not less
Topic Authority Packs — one pillar article plus 4-6 supporting articles on adjacent subtopics, all internally linked — were a strong SEO play in 2020. In 2026 they’re close to mandatory for any topic you want to actually own. The cluster signals topical depth to both Google’s ranking algorithms and to LLM training pipelines.
2. Schema markup is no longer optional
A properly schema-marked page is parseable. An unmarked page requires the algorithm to infer structure. In a compressed SERP where only 3-5 sources get cited, being parseable is a tiebreaker. The most useful schema types for small-business sites: Article, FAQPage, HowTo, LocalBusiness, Service, Organization. Setup is a one-time job per page type; once your CMS template applies the schema, every new page inherits it.
3. Authority signals matter at the brand level, not just the page level
LLMs cite domains they’ve seen referenced in their training data. New domains have a cold-start problem regardless of how good their individual pages are. The remedy: get cited. Industry publications, expert quotes in trade press, podcast appearances, conference talks, contributor pieces. The bar isn’t fame; the bar is being citable enough that other people quote you.
4. Q&A format wins
Pages structured as “here is a specific question, here is the answer with supporting evidence” are dramatically more extractable than essay-form prose. An H2 reading “Does Plan A include feature X?” with a paragraph below starting “Yes — Plan A includes feature X because...” gets cited. A paragraph that buries the same answer in narrative does not.
5. Fresh updates compound
Pages updated quarterly with new evidence and links outrank static pages on the same topic. The bar for “update” isn’t a rewrite; it’s adding new data points, new examples, new internal links to recently published cluster articles. Most CMS platforms support a “last updated” date field; populating it honestly is part of the work.
What still doesn’t work (despite vendors selling it)
- Mass-generated AI content.Easy to detect, easy to deprioritize. Google’s helpful content updates explicitly target low-effort AI-spam pages. LLMs also avoid citing pages whose content patterns suggest mass-generation.
- Keyword stuffing in 2008 form.“Best CRM for small business 2026 · small business CRM 2026 · CRM small business” in the first paragraph — this used to work and hasn’t for over a decade. Still sold by some agencies. Still doesn’t work.
- Buying backlinks from link farms. Manipulative link patterns get algorithmically penalized. Periodic Google manual actions still take out domains that bought low-quality links.
- “Optimization” that ignores user intent.If your page ranks for a query but doesn’t answer the question the searcher actually has, click-through rate signals depress your ranking within weeks.
The honest scope
Modern SEO is a 12-24 month investment. Pages published this month bend the curve six to nine months from now. Topic clusters published this quarter compound for years if maintained. Anyone selling you “ranking in 30 days” is selling you either paid ads (legitimately fast but a different channel) or black-hat tactics (which don’t survive Google’s next algorithm update).
The honest economics: a Topic Authority Pack of 5 articles, well-researched and cluster-linked, costs roughly $1,500-$5,000 to produce and another $200-500/year to maintain. Across a 3-year horizon it’s typically the cheapest source of qualified leads per dollar in your marketing mix. We sell exactly this as a $499 addon (you can see it on our pricing page); we’re publishing three for ourselves as the proof.
The honest closing
SEO didn’t die when AI Overviews arrived. The bar got higher and the rewards got more concentrated at the top. The work that always mattered — substantive content, real expertise, proper structure, earned authority — matters more now, not less.
What changed is the cost of cutting corners. In 2020 you could publish a mediocre article and pick up some long-tail traffic. In 2026 that same article gets bypassed by the AI Overview and receives nothing. The lazy middle is gone. There’s very good and there’s invisible.