If you've felt the shift in your analytics — flat or declining sessions on informational pages, fewer impressions but stable conversions, branded search creeping up — you're not imagining it. The mechanics of how people find brands have quietly changed, and most teams are still measuring against the old model.
What actually changed
Search used to be a retrieval system. You gave Google keywords; it gave you links. Generative search is a synthesis system. It reads the web on your behalf, drafts an answer, and decides which sources to credit. That's a fundamentally different game, because the destination of a query is no longer your page — it's the model's response.
The practical effect is a split funnel. High-intent transactional searches ("emergency plumber Dallas", "best CRM for SaaS") still drive clicks because users want a contact, a price, or a comparison. Informational queries ("what is GEO", "how does Local SEO work") increasingly resolve inside the AI answer itself.
The new discovery funnel
Discovery now happens in two layers. The upper layer is generative — a user asks an LLM and forms an opinion before they ever see your site. The lower layer is conventional — they search you by name, click a citation, or land via a comparison page. The brands that win optimize both layers, not just the second one.
Layer 1: Pre-click conviction
By the time someone reaches your site through AI, they often already trust you. The model has named you, contextualized you, and implicitly endorsed you. That's why AI-referred traffic converts so much better than cold organic — the qualification happened upstream.
Layer 2: Conversion-grade pages
The pages that earn the click still have to do the work: clear positioning, proof, pricing transparency where possible, and a single dominant call to action. Pre-click conviction without a sharp landing page is a leak.
What still works (and why)
It's tempting to throw out the SEO playbook. Don't. The fundamentals that built crawlable, semantically clean sites are exactly what makes a site retrievable by AI. The signals just compound differently.
- Topical depth — comprehensive coverage of a subject still wins, but the unit of measurement is the entity, not the keyword.
- Structured data — schema.org markup is now training-grade context for retrieval models.
- Internal linking — a clean hierarchy helps models reason about which page answers which sub-question.
- Original data and frameworks — LLMs disproportionately cite primary sources and named methodologies.
What stopped working
Thin programmatic pages, exact-match keyword stuffing, and link velocity plays — never great strategies — are now actively harmful. AI retrieval models are sensitive to repetition and template patterns; they downgrade sources that look algorithmic.
Generating 500 location pages with swapped city names used to print money. Today, models cluster those pages as a single low-information entity and almost never cite them.
The citation economy
Mentions are the new links. When Perplexity cites you in an answer, it's the modern equivalent of a high-authority backlink — except it also exposes you to the user at the moment of decision. Tracking citation rate (how often you appear in answers for queries you care about) is becoming a core KPI.
- Rank for the keyword
- Win the click
- Optimize the page
- Build links
- Measure sessions
- Be the cited source
- Win the mention
- Optimize the entity
- Build references
- Measure influence
What to do in the next 90 days
- 1Audit your top 50 informational pages. Identify which queries now have AI Overviews and measure CTR delta.
- 2Pick five flagship topics where you have genuine authority. Consolidate thin pages into definitive resources.
- 3Add or correct schema.org markup — Organization, Product, FAQPage, HowTo. This is your entity grammar.
- 4Publish primary research, frameworks, or benchmarks at least once a quarter. These earn citations disproportionately.
- 5Track citation rate weekly across ChatGPT, Gemini, and Perplexity for 20 priority queries.
The uncomfortable truth
Most companies will spend 2026 fighting the last war — chasing keyword rankings while their share of voice inside AI answers quietly erodes. The teams that adapt early are buying compounding advantage, because models are slow to update their priors. Once you're the canonical source, you tend to stay the canonical source.
The good news: this isn't a different sport. It's the same sport with a different scoring system. Brands that have always invested in clear positioning, original thinking, and technical hygiene are well positioned. Brands that built their organic moat on volume are not.
Frequently asked questions
Is SEO dead?
No. Organic search still drives the majority of high-intent traffic for most B2B and local categories. What's dying is undifferentiated, volume-based SEO that competes purely on keyword coverage.
How do I track AI search visibility?
Track citation rate across ChatGPT, Gemini, and Perplexity for a defined set of priority queries. Tools like Profound, Otterly, and Peec.ai automate this; manual monthly audits work if budget is tight.
Should I block AI crawlers?
Almost never. Blocking GPTBot, Google-Extended, or PerplexityBot removes you from the answer set entirely. The trade-off — protecting content from training — almost never outweighs the lost discoverability.
How long until AI search dominates?
It already does for informational queries in many categories. Transactional and local search still skew classic. Expect a 70/30 informational split toward AI answers by late 2026.
Want this strategy applied to your business?
Get a focused growth audit from our team — we'll map the highest-ROI opportunities specific to your category.