The SEO fundamentals
incumbents skip.
The next customer asking ChatGPT for the best Ford dealer in their town, the best restaurant open Sunday, or the best landscaper before snow flies isn’t typing it into Google. They’re asking an AI assistant. The mechanics that decide whether an assistant cites your site are the same fundamentals that always decided rank on Google — fast load, server-rendered, real content, no tracker bloat — and most local-business websites still skip them.
Google says there’s no “GEO.”
They’re right, and DARKHORSE was building toward that all along.
In May 2026 Google published an article addressing the “Generative Engine Optimization” category head-on. Four explicit statements:
- “Optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” There is no GEO as a distinct discipline.
llms.txtis not special — “you don’t need to create new machine-readable files, AI text files, markup, or Markdown to appear in generative AI search.”- Structured data is not required for AI search — “there’s no special schema.org markup you need to add.”
- “Chunking” content for AI is nothing — not a real optimization.
Google’s article dismantled the four pillars of the GEO industry’s positioning. DARKHORSE doesn’t sell “GEO products.” DARKHORSE ships the SEO fundamentals Google actually rewards — page experience, server-rendered HTML, real content, no third-party tracker bloat, dealer-owned code — while the incumbents charge $1,500/month and ship sites with 40-PSI scores.
What follows is the technical detail. Two honest caveats: (1) Google’s article addresses Google’s AI search specifically; ChatGPT, Perplexity, and Claude each have their own retrieval mechanics, which is why the per-engine breakdown below still matters. (2) Schema isn’t a GEO wedge, but it still helps rich results and Knowledge Graph identity — we include it for that, not for magic AI lift.
Search has already moved.
Most local-business sites haven’t.
This isn’t a future trend. The numbers are already what they are. Four load-bearing facts that should be reshaping every local-business marketing decision right now.
Zero-click searches
Of Google searches now end without a click — the user gets the answer from AIO or AI Mode and moves on.
AI conversion rate
AI search visitors convert at 4.4× the rate of traditional organic traffic when they do click through.
Top-10 organic citation collapse
Of AI Overview citations now come from pages ranking organic top-10 — down from 76% one year earlier. Old SEO no longer protects.
Buyers using AI to decide
Of consumers now use generative AI to research before purchasing. ChatGPT alone holds 68% of AI-assistant share.
AI bots don’t run JavaScript.
SSR matters — that part hasn’t changed.
GPTBot, ClaudeBot, PerplexityBot, and ChatGPT’s OAI-SearchBot all do the same thing: fetch raw HTML, take what’s there, and leave. An analysis of 500 million GPTBot fetches documented zero JavaScript execution. Not partial — zero.
The vendor stack most local businesses run on is built as a JavaScript single-page app (or a slow CMS hauling a megabyte of plugins). Inventory, hours, menu, pricing, even the page’s own headlines hydrate after the bot has already left. By the time the page is “rendered” in a normal browser, the AI bot has moved on. Whatever was supposed to be visible to AI never was.
A DARKHORSE site is server-rendered. The full page — copy, schema, prices, listings — is in the HTML before the page paints. You can prove it in three seconds: open a terminal and run curl https://your-site.com. What you see is what every AI bot sees. With most vendor stacks, what you see is an empty shell. With DARKHORSE, what you see is the page.
There’s also a timing budget. AI crawlers operate on 1–5 second timeouts with fixed compute budgets. Every 100ms above 200ms TTFB costs roughly 3–5% of your effective crawl budget. The typical vendor site averages 1.2 seconds TTFB — losing about a third of its AI visibility before content is even considered. Pages with First Contentful Paint under 0.4 seconds earn 6.7 ChatGPT citations on average; slower pages earn 2.1.
You can’t beat Wikipedia on Google.
You can win ChatGPT in 90 days.
Different AI search engines have wildly different ranking patterns, and that asymmetry is the opportunity most local businesses are missing.
Google AI Overviews are rich-get-richer. 43% of citations go back to Google’s own properties. Wikipedia, major news, and national aggregators dominate. Only 38% of currently-cited AI Overview pages even rank in the organic top 10 — down from 76% one year ago. A single local restaurant, dealer, or contractor can’t beat AutoTrader, Yelp, TripAdvisor, Angi, or Wikipedia on Google AI Overviews this year. That’s a fight for year three and four.
ChatGPT and Perplexity are still meritocratic. For local business queries, 58% of ChatGPT’s cited sources are the business’s own website. ChatGPT pulls local data heavily from Bing Places. Perplexity weights FAQ schema and freshness — its median citation age is 32.5 days versus Google’s 108. A local business with a server-rendered site, comprehensive schema, recent content updates, and a complete Bing Places profile can win ChatGPT and Perplexity citations in 90 days.
That’s the wedge. The vendor stack can’t deliver it because the architecture is the wrong shape — slow, tracker-heavy, schema-thin, content-light. A DARKHORSE site can. Not because we’re magic; because we built it right.
Six things the incumbents skip.
Each one is verifiable. None are toggleable services a vendor can sell you on top of a slow site. Google’s article reframes these as “SEO” rather than “GEO” — the substance is the same; the marketing category Google just dismantled is what changes.
- 01
Server-rendered HTML
AI bots do not execute JavaScript. Your full page must be in the HTML on first response. Verify:
curl https://your-site.com— if you see the content, AI sees it. If you see an empty shell, AI sees an empty shell. - 02
Sub-200ms TTFB
AI crawlers operate on 1–5 second timeouts. Every 100ms over 200ms TTFB costs 3–5% of crawl budget. Verify: PageSpeed Insights, WebPageTest, or browser DevTools Network tab.
- 03
Schema-content parity
Schema markup must describe content that’s actually on the page. Schema claims that contradict visible HTML are flagged as “spammy structured data” and discounted. Bolting JSON-LD onto a thin page produced −4.6% AIO lift in a 2026 Ahrefs study of 1,885 pages.
- 04
FAQ schema for Perplexity
Perplexity weights FAQ/Q&A structure heavily. Pages with FAQPage schema appear in Google AI Overviews 3.2× more often and have 28% higher citation rates across major engines.
- 05
Fresh
dateModifiedPerplexity’s median citation age is 32.5 days; ChatGPT updates 76% of its top citations within 30 days. Required:
datePublished+dateModifiedin JSON-LD, accurate<lastmod>in sitemap, visible on-page date. Stamp-bumping without content changes gets ignored. - 06
AI-aware
robots.txt79% of top news sites accidentally block
Claude-SearchBotwith overly broad rules. Cloudflare blocksGPTBot,ClaudeBot, andPerplexityBotby default on several plans. Correct policy distinguishes training bots from search bots and explicitly allows the ones that drive citations.
Vendors sold hype.
Google just confirmed it.
Every vendor in every vertical is selling a GEO product right now. Dealer.com sells GEO Packages. DealerOn launched OnPrompt. Yext sells Yext AI Search. BirdEye markets AI Reputation Management. Squarespace and Wix added AI-features tabs. The pitch is universal: pay extra, on top of your existing platform, and they’ll make you visible in ChatGPT and Google AI answers.
In May 2026 Google’s own statement gutted the pitch: there is no GEO. llms.txt is not special. Schema markup isn’t required for AI search. “Chunking” content is nothing. The four pillars the GEO industry was selling are the four things Google explicitly said you don’t need.
A 2026 Ahrefs study found the same thing empirically. They added JSON-LD schema to 1,885 thin pages and compared against 4,000 controls. Result: −4.6% in Google AI Overviews. +2.4% in AI Mode. +2.2% in ChatGPT. Schema on a thin page produces near-zero citation lift. The vendors selling GEO products are the same vendors who built the slow, tracker-heavy, schema-thin platforms. Selling a fix for it is the second revenue line, not a solution.
DARKHORSE doesn’t have a GEO product because we ship the SEO fundamentals Google said matter from day one. 95–100 PageSpeed. Server-rendered HTML. Real text content. Zero third-party trackers. Dealer-owned code. None of it is an add-on. All of it is the build.
Six engines. Six rules.
One architecture that works for all of them.
AI search isn’t one channel. Every major engine crawls and cites differently — and a business that optimizes for one accidentally optimizes for none. The DARKHORSE build hits the shared substrate: server-rendered HTML, complete schema, fast TTFB, first-party UI.
OAI-SearchBot + GPTBot
For local queries, 58% of cited sources are the business’s own website. Pulls local data heavily from Bing Places. Yelp, Facebook, and Google Maps do not appear as directory sources.
SSR HTML + schema + Bing Places parity
PerplexityBot
Weights FAQ/Q&A structure aggressively. Median citation age 32.5 days versus Google’s 108. Cites every factual claim with numbered footnotes.
FAQ schema on every page + visible & structured dateModified
5-stage pipeline
Entity-density-weighted (15+ Knowledge Graph entities per 1k words = 4.8× selection). 43% of citations go to Google’s own properties. Hard for a single local business to win head-on.
Comprehensive entity sameAs + Organization schema
Bingbot + Copilot
Bing Places is the foundational local entity. Bing Webmaster Tools publishes an AI Performance Report showing which Copilot queries cite you. Two-for-one with ChatGPT’s Bing reliance.
Bing Places + Webmaster Tools setup in onboarding
Applebot + Maps/Yelp
Applebot crawls; Applebot-Extended is a directive flag (not a crawler). Local results pull from Apple Maps + Yelp partnership. Surfaces in Siri/Spotlight without public citations.
NAP consistency across Apple Maps + Yelp profiles
ClaudeBot + Claude-SearchBot
79% of top news sites accidentally block Claude-SearchBot with overly broad robots.txt — the most common own-goal. Anthropic ToS requires displaying citations when surfacing Claude’s web-search output.
robots.txt that distinguishes training bots from search bots
AI agents are about to shop on customers’ behalf.
Your site needs to be readable by them.
OpenAI shut down Operator and replaced it with ChatGPT Agent in July 2025 specifically because the original failed on JavaScript-heavy flows, CAPTCHAs, and session bloat. Three commerce protocols are now live or in pilot: Agentic Commerce Protocol (OpenAI + Stripe, September 2025), NLWeb (Microsoft, May 2025), and Universal Commerce Protocol (Google + Shopify + Target + Wayfair, January 2026).
None of these protocols can read a JavaScript-hydrated vendor template. Agentic shoppers need deterministic structure: one URL per product or listing, server-rendered listing data, HTML form fallbacks for lead capture, and no third-party widget bloat that breaks the automation. Every DARKHORSE site ships agent-ready by default:
- SSR-rendered Product / Service / VehicleListing JSON-LD — agent reads the structured data first, the visible page second; both match
- Deterministic canonical URLs — one item = one URL, no session-state in paths
- HTML form fallbacks — lead capture works without JavaScript so an agent can submit on a customer’s behalf
- First-party chat and UI — no third-party widget injection surface for the agent to mis-handle or get hijacked through
- ImageObject schema on every image — multimodal-search-ready, with descriptive alt text not
alt="image" - Schema-content parity — what the agent reads in JSON-LD matches what the agent renders in HTML, every time
The vendor templates fail at least four of these six today. They’ll be retrofitted eventually — but every business paying for the upgrade is paying for what should have been the foundation.
Every vendor widget is a place hostile content can hijack the AI shopping for your customer.
Google’s Security Research, Brave’s browser team, and Palo Alto’s Unit 42 have all documented indirect prompt injection in the wild against AI browsers and chatbot plugins through 2025–2026. The technique: malicious instructions hidden in 1-pixel text, zero-opacity color, or off-screen positioning — often delivered through embedded third-party widgets, analytics tags, or chat plugins. Documented attempts rose 32% from Nov 2025 to Feb 2026.
For any local business site, the practical impact is the same: every vendor chat widget, embedded form, third-party trust badge, and analytics tracker is a separate attack surface an attacker can use to manipulate an AI agent that’s helping a customer shop. The agent reads the page, ingests the malicious instructions hidden in the widget, and acts on them — mis-quoting prices, mis-citing inventory, routing the shopper elsewhere. The business’s brand pays for the failure.
First-party UI by default. Every DARKHORSE site ships with the chat widget, lead-capture forms, calculators, and trust signals built natively into the application — no third-party scripts, no embedded iframes, no injection surface. The site has nothing to hand a regulator and nothing to hand a hostile prompt.
Most local-business website platforms depend on third-party scripts, widgets, and vendor tag stacks for revenue. DARKHORSE is built differently: first-party UI, zero tracker bloat, and ownable code from day one.
We build the site.
You still have to win the off-site battle.
Off-site signals — Google Business Profile completeness, review volume and response cadence, NAP consistency across the directories your customers use — are roughly 70% of AI ranking weight for local queries. DARKHORSE doesn’t manage any of that today. The site is the foundation; the off-site signal is the building. We tell you what matters even when we don’t do it.
What we ship
- 95–100 PageSpeed across all four categories, mobile
- Server-rendered HTML with comprehensive Schema.org markup
- Sub-200ms TTFB on every page
- Zero third-party trackers, zero session recorders, zero data brokers
- FAQ + Product/Service + Organization schema on every relevant page
- Full admin panel — you edit content yourself in 30 seconds
- robots.txt that distinguishes training bots from search bots
- Entity sameAs to your Google Business Profile + vertical directories
- Fresh
dateModifiedon content that actually changes
What you still own
- Google Business Profile — weekly photos, monthly Google Posts, hours accuracy
- Review volume — 200+ Google reviews, 30+ new in the last 90 days
- Response cadence — 80%+ response rate within 48 hours
- NAP parity across the directories that matter for your vertical (Edmunds/KBB/Cars.com for auto; Yelp/OpenTable for restaurants; Angi/HomeAdvisor for services)
- Bing Places (ChatGPT pulls from this) — many businesses haven’t even claimed theirs
- BBB and Chamber of Commerce profiles
- For multi-location operators: separate GBP per location, no duplicated content
If you want help auditing your off-site signal — what the AI engines actually see when they look at your business today — send the URL. We’ll send back a free dossier with the four AI ranking signals mapped to your current state, plus the gaps. That part we do, free.
Same architecture.
Different context.
The AI search mechanics above apply to every local business. The vendor traps, the off-site directories, the regulatory exposure, and the specific shopper-query patterns vary by industry. Pages with industry-specific framing:
For Car Dealers
OEM co-op, approved-vendor lists, DMS integration (DealerBuilt, Tekion), FTC pricing enforcement, GLBA Safeguards Rule for dealers as financial institutions.
For Restaurants & Groups
Menu schema (the #1 cause of restaurant invisibility), OpenTable / Resy vendor lock, multi-location umbrella positioning, seasonal swing management.
For Home Services
Service-area schema, before/after photo galleries with ImageObject schema, Angi / HomeAdvisor lead-share alternative, seasonal demand patterns.
Questions about AI search.
Answered.
Will AI assistants like ChatGPT and Perplexity cite my business?
Yes, if the architecture supports it. AI engines cite server-rendered HTML with structured data and verifiable trust signals. For local business queries, 58% of ChatGPT's cited sources are the business's own website. Perplexity's median citation age is 32.5 days versus Google's 108, so freshness matters too. Off-site signals (your Google Business Profile completeness and review volume) still matter for the local ranking layer — DARKHORSE doesn't manage those, but the site we build is one of the four conditions AI weighs.
Do I need a GEO product or service?
No. In May 2026 Google explicitly stated there is no separate "GEO" discipline — optimizing for AI search is just SEO. They specifically said you don't need llms.txt, you don't need special schema markup, and content "chunking" for AI is not a real optimization. The vendors selling GEO products (Dealer.com GEO Packages, DealerOn OnPrompt, Yext AI Search, BirdEye AI Reputation Management) are mostly selling add-ons to sites that fail SEO fundamentals — slow load, JavaScript-heavy frontends, third-party tracker bloat, vendor-locked code. A 2026 Ahrefs study found schema added to thin pages produced near-zero AI citation lift (−4.6% in Google AI Overviews). DARKHORSE doesn't sell a GEO product because the SEO fundamentals Google said matter are baked into every build from day one.
Does my current website let AI bots see it?
It depends on three things. First, your site needs to be server-rendered: AI bots like GPTBot, ClaudeBot, PerplexityBot, and OAI-SearchBot do not execute JavaScript, so a JS-hydrated single-page app is invisible to them. Second, your time-to-first-byte needs to be under about 200 milliseconds: AI crawlers operate on 1–5 second timeouts. Third, your robots.txt needs to permit them: 79% of top news sites accidentally block Claude-SearchBot through overly broad rules, and Cloudflare blocks GPTBot, ClaudeBot, and PerplexityBot by default on several plans. You can test the first two yourself in a terminal: `curl https://your-site.com`. If you see your full page, AI bots see it. If you see an empty shell, they see an empty shell.
What is the difference between SEO and GEO?
Per Google's May 2026 statement: there isn't one. "Optimizing for generative AI search is optimizing for the search experience, and thus still SEO." The GEO category was largely a marketing invention by SEO vendors looking for an upsell. The substance — fast load, real text content, server-rendered HTML, no tracker bloat, ownable code — is the same SEO fundamentals that always mattered. What's true: AI answer engines reward content patterns slightly differently than ten-blue-links search (answer-first passages, citation-ready facts, recency signals). What's also true: those are still SEO best practices, not a separate discipline that justifies a $1,500/month "GEO package" bolted onto a broken site.
How long does it take to start ranking in AI search?
Faster than traditional SEO, but not instant. Princeton's 2024 GEO research and ongoing 2025–2026 industry studies suggest 30–90 days from a clean architectural rebuild for ChatGPT and Perplexity citations to start appearing for local queries. Google AI Overviews take longer (3–6 months) because they weight third-party authority signals that compound over years. Off-site work — Google Business Profile completeness, review velocity, NAP consistency — compounds on its own timeline, typically requiring a full year before AI engines treat it as established signal. A dealer or restaurant that started building citation graph in 2023 has a 3-year head start an LLM has been crawling continuously.
Where the numbers come from.
The statistics on this page draw on public AI-search research published 2024–2026. Direct citations to specific firms are kept inline where the original study is identifiable (the 2026 Ahrefs schema-thin study of 1,885 pages, Google’s May 2026 guidance). The remaining numbers come from a body of ongoing industry research; the source families that publish this work include:
- Google Search Central — May 2026 guidance on optimizing for generative AI features (developers.google.com/search/docs/appearance/ai-features).
- Ahrefs — 2026 study on schema markup and AI Overview citation lift (cited inline).
- SparkToro (Rand Fishkin) — zero-click search behavior research, 2024–2025.
- Industry AI-search benchmark publishers — Profound, Semrush, BrightEdge, SE Ranking, Authoritas. Engine-specific citation patterns, freshness windows, and own-site citation ratios.
- AI-crawler access and behavior — Originality.ai, TollBit, Cloudflare, Vercel reports on bot timeouts, robots.txt patterns, and platform-level access.
- Princeton GEO research (2024) and follow-on industry studies on citation-graph compounding.
AI search behavior changes quickly. Numbers here are point-in-time benchmarks current as of 2026-05-19. Directional claims (faster sites earn more citations, schema on thin pages doesn’t lift, server-side rendering is required for AI bots) are more durable than specific ratios. If you need any single statistic audited to its originating study before quoting it externally, ask — we’ll walk through the source.
Skip the scan.
Just talk.
Every business’s AI-visibility situation is different. We can run the numbers on your specific stack — what an AI bot actually sees when it fetches your site, what your competitors are doing, what changes if the site renders server-side and ships with zero third-party trackers.
Or scan your current site first — up to you.