GEO fundamentals¶
Generative engine optimization is the work of becoming the source an AI assistant cites when someone asks a category question — not "find MyProduct" but "what's the best way to solve my problem". The fundamentals are: reframe from brand to intent, understand the three surfaces you're optimizing at once, replace folklore with the measured ground truth, and spend effort down a prioritized lever list that starts with crawlability. This chapter is that foundation; the rest of the part is depth on each lever.
The reframe: branded discovery is easy, intent discovery is the game¶
Ask ChatGPT "what is customdomain.ai" and — once the site is crawlable and indexed — it will answer. That's branded discovery, and it's table stakes. The valuable, hard problem is intent discovery: being the answer when the asker doesn't know you exist.
- "How do I let users use their own domain on my SaaS?"
- "How much does snow removal cost in Shakopee?"
- "Best tool to automate TLS for customer domains"
Nobody typing those queries knows your name. Winning them is what GEO is for — and it's measurably distinct work. On customdomain.ai we found the branded query literally returned zero results (Google tokenized the dotted brand into generic words — see the tokenization crisis) while intent queries were winnable in weeks because their SERPs were vacant (measured, 2026-07).
The reframe has a second half. A product now has two users pulling different levers:
- The human buyer, who Googles or asks ChatGPT. Won with everything in this part: crawlability, Bing, answer-first content, schema, off-site citations.
- The AI agent, told "add custom domains to the app you just built." It doesn't read your blog — it consults registries and
/.well-known/endpoints. That's the AI Agents part, and for a developer-facing product it's the greenfield bet.
Keep the split explicit in your planning: findability (marketing surface) and agent-usability (API/MCP surface) are different codebases, different owners, one product story.
The three-surface model¶
Every intent query can now resolve on three surfaces, and they share plumbing:
| Surface | Who answers | What feeds it | Your lever |
|---|---|---|---|
| Classic search | Google, Bing result pages | Google/Bing indexes | Crawlability, content, links — Google, Bing |
| AI answers | ChatGPT Search, Perplexity, AI Overviews/AI Mode, Claude, Copilot | Bing (ChatGPT, Copilot), Google (AI Overviews), Perplexity's own crawl, plus each vendor's on-demand fetchers | Everything in this part |
| AI agents | Agents acting for a user | MCP registries, /.well-known/*, tool descriptions, docs |
AI Agents |
The non-obvious dependency: answer engines sit on search indexes. ChatGPT's search retrieval correlates ~87% with Bing's top organic results (external research, compiled 2026-07); AI Overviews draw on Google's index. So classic SEO didn't die — it became the substrate. A page that ranks nowhere on either index can still be cited only via the third route, a user pasting your URL — which you can encourage but not rely on (the Ask-AI widget).
Ground truth: the facts that kill the myths¶
Most GEO advice online is hype. These are the load-bearing facts, each dated and attributed. When a vendor pitch contradicts one of them, ask for their evidence.
- Crawlability is lever #1 — and it fails silently. A WAF or bot-challenge mode can return 429/403 to every crawler while humans see a normal site. Measured on headsupoutdoorservices.com, July 2026: Vercel's Attack Challenge Mode challenged Googlebot, GPTBot, and PerplexityBot alike — including on
robots.txtandsitemap.xml, so crawlers couldn't even read the crawl directives. Result:site:showed zero pages and the business's Facebook page outranked its own domain. Full diagnosis in Rendering, WAFs, and bot challenges. - Bing gates ChatGPT. ChatGPT search runs on Bing; ~87% of its citations match Bing top-organic results (external research, compiled 2026-07). Google rank does not drive ChatGPT citations. Verify your site in Bing Webmaster Tools and submit the sitemap there, not just in Search Console. In our 2026-07-15 portfolio scorecard, only one of seven live properties had done this (measured).
- llms.txt is largely decoration for AI chat. Ahrefs' server-log study across ~137k domains found ~97% of llms.txt files get zero bot requests; Google's Gary Illyes stated (2025-07-23) that Google does not use llms.txt for Search, AI Overviews, or AI Mode. Its one real consumer is coding agents reading docs sites. The honest chapter.
- Structure gets extracted; prose gets skimmed. HTML comparison tables are extracted ~81% vs ~23% for the same facts as prose; FAQ-structured pages earn ~3x more citations; ~44% of ChatGPT citations come from the first 30% of a page (external research, compiled 2026-07). Hence the answer-first recipe in Content that gets cited.
- Adding evidence lifts citations. The Princeton GEO study (Aggarwal et al., 2023–24) measured +40% generative-engine visibility from adding statistics, +40% from citing authoritative sources, +28% from quotations — versus ~0% for keyword stuffing (external research).
- Rich schema correlates with citation. ~61% of ChatGPT-cited pages carry rich schema vs ~25% of the baseline web; three or more schema types per page correlates with more citations (external research, compiled 2026-07).
- Most citations are third-party. ~85–93% of AI citations point somewhere other than the brand's own site. Reddit is the #1 cited source (~40% overall; ~47% for Perplexity); G2/Capterra are top-cited for B2B SaaS; only ~11% of domains get cited by both ChatGPT and Perplexity, because the engines source differently (external research, compiled 2026-07). → Off-site signals
- There is no force-index button. Google's Indexing API is officially JobPosting/BroadcastEvent-only — off-label use risks revocation (documented). The sitemap-ping endpoint was removed in 2023 and now 404s (documented). Search Console's "Request Indexing" is a rate-limited hint (~10–12 URLs/day observed, unofficial). IndexNow is real but feeds Bing/Yandex and friends — not Google.
- Google's own guidance names the myths. Google's AI-search documentation explicitly lists AI-specific rewriting, fragmenting content into tiny chunks, and special "AI markup" as ineffective (documented). There is no secret markup; there is structure, evidence, and retrievability.
- Freshness is engine-specific. Perplexity favors recent content and recrawls fresh pages in ~2–7 days (external research, compiled 2026-07); Google refreshes on its own sitemap-recrawl schedule; training corpora refresh on model releases. Publish ahead of seasonal demand for Perplexity; keep
<lastmod>truthful for everyone (Content that gets cited).
The prioritized lever list¶
Spend in this order. Each lever is cheap to verify before you invest in the next.
| # | Lever | Effort | Failure mode if skipped | Chapter |
|---|---|---|---|---|
| 1 | Crawlability — robots allows AI bots; no WAF challenge; server-rendered HTML | Minutes to check | Everything else is invisible | AI crawlers |
| 2 | Bing pipe — Bing Webmaster verification, sitemap submitted, IndexNow on publish | ~1 hour | ChatGPT/Copilot can't retrieve you | Bing & Beyond |
| 3 | Answer-first content for intent queries — pillar + cluster, BLUF, tables, question H2s | Days–weeks | Retrievable but never quoted | Content that gets cited |
| 4 | Schema on content pages — FAQPage, DefinedTerm, TechArticle + the site graph | Hours per page type | Weaker extraction and entity confusion | Structured data |
| 5 | Off-site presence — reviews, Reddit/forums, directories, comparisons, GitHub | Ongoing, slow | Capped ceiling: engines cite third parties ~85–93% of the time | Off-site signals |
| 6 | Entity clarity — Organization + sameAs graph, consistent naming, Wikidata | Hours | Engines can't verify you exist; misattribution | Entities and trust |
| 7 | Measurement loop — the query battery, citation logging, re-audit cadence | ~1 hour/month | You can't tell what worked | AI visibility audit |
The ordering logic: levers 1–2 gate everything and cost almost nothing to verify, so verify them first, personally — curl as the bots, don't assume. On the headsup engagement, the very first act was fetching the site as Googlebot/GPTBot/PerplexityBot; that one check reframed the entire program, because no amount of content work would have mattered through a 429 wall (measured, 2026-07).
What NOT to buy¶
The snake-oil catalog, with reasons:
- "Instant indexing" tools — repackaged off-label Indexing API use (revocation risk) or dead sitemap pings. No legitimate general-purpose force-index exists (documented).
- llms.txt packages sold as "AI SEO" — see fact 3. Ship the cheap version for docs if you like (decision rule); never pay for it as a citation lever.
- "AI-optimized rewriting" services — Google names AI-specific rewriting and tiny-chunk fragmentation as myths (documented). Structure and evidence work; dialect-for-robots doesn't.
- Guaranteed mentions / paid citations — paid placement is named ineffective in Google's guidance, and astroturfed community presence gets deleted and reputationally backfires (Off-site signals).
- Fake reviews or invented aggregate ratings — a policy violation with a manual-action risk (Google's Dec-2025 review policy — Reviews), and an authenticity failure engines increasingly detect.
- "OKF" (Google Open Knowledge Format) as an SEO play — it's a 2026 Google Cloud spec for feeding your own internal agents knowledge; it is unrelated to indexing or external discovery (documented).
Verify it yourself¶
Don't take this chapter's word for your own site. The 15-minute check:
# 1. Can the AI bots actually fetch you? (200 = yes; 403/429 = you're invisible)
for ua in "GPTBot" "OAI-SearchBot" "ClaudeBot" "PerplexityBot" "bingbot"; do
echo "== $ua =="
curl -s -o /dev/null -w '%{http_code}\n' -A "$ua" https://YOURSITE.com/
done
# 2. Does robots.txt allow them? Does the sitemap advertise the right host?
curl -s https://YOURSITE.com/robots.txt
curl -s https://YOURSITE.com/sitemap.xml | head -20
# 3. Is your content in the HTML at all, or client-rendered?
curl -s https://YOURSITE.com/ | grep -i "a distinctive phrase from your homepage"
Then ask ChatGPT and Perplexity three of your intent queries and note who gets cited — that's your displacement target and your baseline (the full audit).
Gotchas¶
- Optimizing only for Google. The reflex of a decade of SEO. Bing gates ChatGPT and Copilot; check Bing Webmaster Tools is verified before polishing another Google detail.
- Trusting tool summaries over raw responses. A fetch-and-summarize tool once reported missing alt text that a raw curl disproved (measured, 2026-07). Verify claims against the actual HTML an engine would see —
view-source, not DevTools' rendered DOM. - Point-in-time audits rot. Three of ten findings in one audit had already been fixed by the time implementation started (measured, 2026-07). Re-verify findings against the live site before acting on them.
- Winning the argument, losing the entity. All the content in the world doesn't help if engines can't resolve who you are — duplicate Organization nodes, inconsistent naming, and a tokenizing brand name each sabotage citation attribution. Fix the entity layer alongside content.
- Measuring nothing. Record your baseline (GSC + Bing + the query battery) before the program starts, or you'll never prove what moved. → Measurement
Related¶
- Content that gets cited — the on-site execution of these fundamentals
- AI crawlers and crawlability — lever #1 in depth
- Off-site signals — the highest-ceiling lever
- How AI finds and cites — the retrieval mechanics
- Keyword and SERP strategy — intent-query research and winnability triage
- Source skill: generative-engine-optimization