Why are small sites losing 60% of organic traffic to AI search but big sites only 22%?

AI Overviews and ChatGPT Search retrieve answers from a small number of well-cited sources. Big publishers have the citation graph, brand mentions, and structured data that AI engines preferentially pull from; small sites get summarized away in the answer itself.

What is the difference between SEO and GEO?

SEO optimizes a page to rank in a list of links. GEO (Generative Engine Optimization) optimizes a page to be cited inside an AI-generated answer. The mechanics differ: GEO rewards extractable definitions, numeric answers, comparison tables, and clear author signals.

What schema actually moves AI citations?

Article + Author (Person), FAQPage, HowTo, and Product+Review are the four that AI engines parse most reliably. Sites that ship all four typically see citation rates rise within 30-60 days even without new traffic.

Should I stop doing traditional SEO?

No — traditional SEO still produces direct clicks for branded and high-intent queries. Drop the parts that no longer pay back: thin top-of-funnel listicles, keyword-stuffed roundups, and generic "ultimate guides." Keep the parts AI engines cite: definitions, frameworks, and original data.

How solo founders fight back when AI search cuts traffic 60%

All posts

In 2025 and 2026, Google AI Overviews, ChatGPT Search, Perplexity, and Gemini rewrote how discovery works. For large publishers, the traffic hit averaged around 22%. For small sites and early-stage SaaS, it averaged closer to 60%. The reason is structural: AI answers now satisfy the query before the user ever clicks, and when a click does happen, the sources it sends to are a curated shortlist - not a blue-link page of 10 competitors. If your content isn't inside that shortlist, your traffic drops no matter how good your SEO was last year.

GEO - Generative Engine Optimization - is the new discipline for getting cited by AI answers instead of replaced by them. It overlaps with SEO but it's not the same job. This is the full playbook for solo founders: how AI search retrieves sources, what content shapes get cited, the schema that still matters, what to drop from your old SEO stack, and a 30-day sprint to start showing up.

Why small sites got hit 60% and big sites got hit 22%

AI answers don't replace all queries equally. The queries that convert to an AI answer cleanly - definitions, comparisons, how-tos, "best X for Y" - are exactly the ones small sites were ranking for. Big publishers' remaining traffic sits on brand queries, news, and reference content that AI engines still link out to. Small sites were living on the bottom 80% of search volume, which is exactly the volume AI Overviews now satisfy without a click.

The other half of the asymmetry: AI engines over-index on trusted, citation-worthy sources. That biases them toward publishers with existing authority, structured content, and schema markup - which most small sites don't have. The fix is not "more SEO." The fix is rebuilding content so AI answers actually want to cite it.

How AI search actually retrieves sources

Every AI answer engine - Google's Gemini-powered AI Overviews, ChatGPT Search, Perplexity, Claude's web tool, Gemini app - follows roughly the same pipeline:

Decompose the query into 2-4 sub-questions the model needs to answer.
Retrieve candidate sources via a search API (Google, Bing, Exa, Brave) and sometimes a vector/embedding search over a curated index.
Score and rank the sources for relevance, trust, recency, and specificity.
Extract usable passages - typically self-contained 2-5 sentence chunks that answer a sub-question directly.
Compose the answer and cite the 3-6 best passages with links.

To be cited, your content needs to (a) be retrievable by the underlying search API in the first place - classic SEO still applies - and (b) contain passages the model can lift cleanly. Long rambling paragraphs with a point buried in sentence seven don't get cited. Clean, self-contained, direct answers do.

The 5 content shapes AI engines cite most

Across AI Overviews, ChatGPT Search, and Perplexity, five content patterns dominate the citation list:

Definition blocks. "X is [one-sentence definition]. It works by [two-sentence mechanism]. It matters because [one-sentence relevance]." 4-5 sentences, self-contained.
Direct answers with numbers. Specific figures, ranges, or benchmarks in the first sentence. AI engines preferentially cite passages that contain numeric specificity.
Comparison tables. Side-by-side matrices (ideally rendered as HTML tables, not images) get lifted verbatim.
Step lists. Numbered procedural lists with one action per step and no embedded subheaders.
FAQ blocks. Question-shaped H2s followed by a 2-4 sentence answer. Each pair gets indexed independently by AI engines and cited in the right sub-query.

Every pillar post on your site should contain at least three of these shapes. Articles that are pure narrative prose get cited at a fraction of the rate of articles with clear extractable blocks.

Schema and structured data that actually move citations

Most schema tutorials are still written for 2018 SEO. For GEO specifically, only a handful of schema types reliably help AI engines identify your passages:

Article + author. Article JSON-LD with a real author reference to a Person object (with sameAs links to LinkedIn, Twitter, etc.) gives AI engines an E-E-A-T signal they can actually verify.
FAQPage. Each Q+A gets treated as a standalone passage. This is the single highest-ROI schema addition for GEO.
HowTo. For step-by-step content. AI answers in "how do I X" queries lift HowTo steps directly.
Product and Review. For comparison/alternative posts - Product schema on each tool plus AggregateRating makes "best X for Y" queries cite you more often.

Skip the schema types that don't matter for GEO: Breadcrumb, WebSite, and Organization are fine but won't move citations. Don't spend time on them before the four above are in place.

What to drop from your old SEO playbook

Several SEO tactics that mattered in 2022 are now neutral or counterproductive in the AI era:

Keyword density. Dead. AI engines retrieve by semantic match, not keyword frequency. Stuffing a keyword 14 times actively hurts you.
2,000-word minimums. A 900-word article with 4 extractable passages beats a 2,500-word article with none. Length for its own sake is noise.
Generic outline templates. The intro-definition-benefits-cta template is so recognizable that AI engines deprioritize it. Open with the specific claim or number, not "In today's fast-paced world..."
Link-farm backlinks. Still harmful, now even more so - AI engines penalize sites with unnatural backlink patterns much harder than 2022 Google did.

Keyword strategy in the GEO era

Don't chase head terms anymore. "best CRM" or "what is CAC" gets fully answered by AI Overviews without a click. Focus instead on long-tail, specific, high-intent queries where the AI still needs to cite because there isn't a canonical answer - things like "best CRM for a 2-person real estate team under $50/month" or "how do I calculate CAC when 70% of leads come from organic."

These are the queries AI engines citate heavily because no single source has a perfect answer - the model needs to synthesize. If you're the source with the specific combination the user asked about, you get cited. Write for the specific combination, not the head term.

Brand and author signals AI engines look for

AI answers heavily weight brand mentions across the web - not backlinks specifically, but mentions. Getting cited by a mid-size publication, mentioned in a popular newsletter, or referenced in a widely-quoted LinkedIn post all push your domain up in the AI engine's trust scoring.

Author signals matter even more. A founder with a LinkedIn profile that cross-links to their byline on their site, with a bio that names specific experience, is treated as a more trustworthy source than an anonymous "Team" byline. Do the author-page work. AI engines actually read it.

How to measure if GEO is working

Traditional SEO metrics don't capture AI citations. The signals that matter now:

AI-referred traffic. GA4 / Plausible / PostHog now show referrers like chatgpt.com, perplexity.ai, gemini.google.com. Segment them. If these grow month-over-month while Google organic is flat, GEO is working.
Citation audits. Monthly: ask ChatGPT Search, Perplexity, and Google AI Overviews your top 10 target queries. Note which sources they cite. You want your domain in the citation list for 3-5 of them within 90 days of publishing a pillar.
Branded AI queries. Ask "who is [your company]" and "what does [your product] do" in each AI engine. Is the answer right? Is your site cited? Gaps here point at missing Organization/FAQ schema.

The 30-day GEO sprint for solo founders

Week 1 - audit and prioritize. List your top 20 target queries. Run each through ChatGPT Search, Perplexity, and Google AI Overviews. Record which ones you appear in, which you don't. Pick the 5 highest-intent queries to fix first.
Week 2 - restructure existing pillars. For each of the 5 target queries, take your best existing post and add a definition block, a numeric-answer passage, and a 5-item FAQ. Ship Article + Author + FAQPage JSON-LD.
Week 3 - write 2 new long-tail pillars. Target very specific combinations ("best X for Y under $Z"). These become the queries AI engines need to cite because no head-term article answers them cleanly.
Week 4 - author + brand signals. Build out the author page with sameAs to LinkedIn/X. Get 2-3 external mentions via guest posts, podcast appearances, or newsletter cross-publishes. Re-run the week-1 citation audit - you should see your domain appear in at least one new AI answer.

Bottom line

The 60% traffic hit small sites took from AI search is permanent - nobody's bringing back the 10 blue links. But the citations AI engines surface are a new kind of traffic, and they favor content built specifically to be cited. GEO is extractable passages, useful schema, specific long-tail queries, and real author signals. Rebuild 5 pillar posts with those four traits, get 2-3 external mentions, and you'll start showing up in answers you weren't showing up in a month ago. The solo founders treating this as the new SEO are the ones who'll own their category's AI answer box in 12 months. The ones still chasing keyword density are losing ground they'll never get back.