GEO in 2026: How to Build a Website AI Can Actually Find

Welcome back to the front row.

Every fourth issue of NVM, we go deep.

Not a roundup. Not three bullets and a take. A proper field guide on one topic, written for practitioners who need to actually do something with it, not just nod along. The regular format returns next week. This is the one you save.

Issue #4 is Deep Dive No. 01. The topic: how to build a website AI can find, read, and cite. nvm.marketing goes live this week. This is the checklist running before it does.

Launching Into Silence

The Complete Field Guide to GEO for New and Existing Websites

nvm.marketing is launching this week.

Like most practitioners, I've done this before. You submit the sitemap. You set up Search Console. You clear the robots.txt. Standard sequence. Works for Google. Takes 48-72 hours for the first crawl. Usually fine.

But this time there's a second checklist running underneath the first one — and it has nothing to do with Google.

Because here's what the data says about the state of search in 2026: only 44.3% of pages that rank in Google's top 10 appear in a single AI-generated answer. The majority don't. The overlap between what Google ranks and what AI engines cite has dropped from 70% to below 20% in 18 months. Ranking is no longer the same as being discovered.

And nobody sends you the second checklist.

This is it.

Why this matters more than it did six months ago

AI search is not a future problem. It's a current traffic source.

AI-referred sessions to websites grew 527% year-over-year through 2025, according to Previsible's AI Traffic Report — from roughly 17,000 to 107,000 monthly sessions in a tracked sample of 19 GA4 properties. More importantly: conversion quality from AI-referred traffic is significantly higher than traditional organic. Users who arrive from a ChatGPT or Perplexity citation are already qualified. The AI already answered their question, recommended your site, and they chose to click. That's not top-of-funnel.

Now add: Gartner projects traditional search engine volume will drop 25% by the end of 2026 as users shift to AI-generated answers. And 31.3% of the US population will use generative AI search in 2026, according to EMARKETER — a number that has MENA equivalents growing faster as mobile-first AI adoption accelerates in the region.

You don't need to panic. You need a framework.

The three-layer stack — in order, not optional

The most important thing to understand about GEO is that it doesn't replace what came before. It layers.

Layer 1 — SEO: the foundation. If Googlebot can't reach your content, neither can GPTBot. AI engines use live-web retrieval for real-time answers. That means crawlability, indexability, page speed, mobile optimization, and clean site architecture all remain prerequisites. Fix technical SEO before touching anything else. This layer is not optional and it is not new.

Layer 2 — AEO (Answer Engine Optimization): the structure. AEO is the work of making your content machine-extractable. Schema markup. FAQ structure. Heading hierarchies where each H2 and H3 functions as a standalone question. Paragraphs that open with a direct answer before expanding with context. AEO turns your content from prose into passages — the unit that AI retrieval actually operates on.

Layer 3 — GEO (Generative Engine Optimization): the citation layer. GEO is where authority, entity recognition, and brand consistency determine whose content gets cited from the pool of technically accessible, well-structured pages. This is where third-party mentions, Reddit presence, YouTube content, and LinkedIn consistency start functioning as ranking signals — not for Google's list, but for an AI engine's answer. GEO is what determines whether your name appears in the response.

The framework from Search Engine Land that I've found most precise: SEO is the foundation, AEO is the frame, GEO is the roof. You cannot build the roof before the frame. The sequence matters.

Phase 1 — Access and crawlability

(Impact timeline: 2–4 weeks)

Before any optimization matters, AI engines need to physically reach your content. This is where most new sites fail silently — and where the fix is fastest.

Check your robots.txt today.

Open yourdomain.com/robots.txt. Look for Disallow rules and check them against this list of AI crawler user agents you want to allow:

• GPTBot — OpenAI (ChatGPT, SearchGPT)
• ClaudeBot — Anthropic (Claude)
• Google-Extended — Google's AI training and products
• PerplexityBot — Perplexity
• Applebot-Extended — Apple AI features
• Bytespider — ByteDance / TikTok AI

If any of these appear under a Disallow: / rule and you haven't put them there intentionally, remove them. Most site migrations and template-based builds include blanket bot restrictions that sweep up AI crawlers. Check this before you optimize a single word of content.

The Cloudflare trap.

This is the highest-frequency silent failure I've seen in MENA-based site setups. Cloudflare updated its default Bot Fight Mode configuration in 2024. Depending on your plan and settings, Cloudflare may now be automatically blocking AI crawler user agents without any notification to you. The crawl appears to be working. Google Search Console shows coverage. But ChatGPT and Perplexity have never seen your site.

To check: open your Cloudflare dashboard → Security → Bots. If Bot Fight Mode is on, review the bot-blocking rules. Look for your server logs and filter for GPTBot and ClaudeBot user agents. If you're returning 403s to those user agents, your GEO work is happening behind a wall.

JavaScript rendering.

AI retrieval systems, like the Googlebot of five years ago, often struggle with JavaScript-rendered content. If your page body — the actual copy an AI would extract as a passage — only appears after JS execution, you're invisible to a significant share of the AI retrieval layer. Server-side rendering or static generation solves this. Check your core content pages: view the page source (Ctrl+U) and verify the body copy is present in the raw HTML, not injected by script.

Phase 2 — Structure and schema

(Impact timeline: 4–8 weeks)

Once your content is accessible, the job is making it extractable.

Schema markup — three types, deployed in order.

Start with FAQPage schema on any page that answers questions — which for a marketing, agency, or SaaS site is most pages. Then Article schema on all long-form content, with explicit author, datePublished, and dateModified attributes. The dateModified field matters significantly: AI systems weight recency, and a clear update timestamp on refreshed content is a citation-priority signal. Finally, BreadcrumbList schema to establish your content hierarchy for AI systems trying to understand how your site is organized.

Use Google's Rich Results Test and Schema.org's validator together — they catch different issues and produce different errors.

Answer-first paragraph structure.

This is the single most impactful content change you can make for GEO, and it costs nothing. Rewrite every section so the opening sentence directly and completely answers the primary question that section addresses — before you expand, qualify, or provide context. AI retrieval systems that use RAG pull from the opening content of pages disproportionately. The first 200 words of an article are your citation pitch. Write them accordingly.

Test this: take any key page on your site. Ask ChatGPT the question that page is supposed to answer. If the AI produces an answer that sounds nothing like your first paragraph, your first paragraph isn't extractable.

Heading hierarchies as question scaffolds.

Your H2 and H3 headings should function as standalone questions that your content directly answers. Not "Overview" or "Background" — "What is GEO and how is it different from SEO?" Every heading is a potential retrieval sub-query. Structure your headings so that someone reading only the heading level understands the full argument of the page. AI systems do this constantly.

Content specificity — the most underestimated GEO factor.

The Princeton KDD research paper that established GEO as a discipline found that combining statistical evidence with fluency optimization — cited numbers plus clear writing — produced a 40% increase in AI visibility. Content with FAQ structure and direct declarative statements showed 30–40% higher citation rates. This translates to a simple rule: every claim should have a number, and every number should have a source. "Performance marketing has changed" doesn't get cited. "AI-powered bidding now drives 78% of all Google Ads spend in 2026" does. Specificity is the difference between a passage an AI can extract and one it skips.

Phase 3 — Content architecture and the prompt map

(Impact timeline: 8–12 weeks for AI citations to stabilize)

Phase 1 gets AI engines to your door. Phase 2 makes your content legible. Phase 3 is about owning the questions in your space before your competitors do.

Build your prompt map with AnswerThePublic.

AnswerThePublic — owned by Neil Patel, integrated with Ubersuggest — pulls real autocomplete data from Google, YouTube, and Bing and organizes it into question clusters: who, what, when, where, why, how, which. The new AI Dashboard added in 2026 also shows how AI engines currently describe your topic — which gives you the framing before you optimize for it.

Here's the GEO-specific way to use it: enter your core topic. Export the question cluster. Filter specifically for the question-format queries — "how do I," "what is," "why does," "when should." These are your prompt targets. They're not keyword phrases — they're the sub-queries AI engines fan out to when synthesizing an answer about your topic. Build one piece of content per question cluster, not per keyword. Each piece should directly answer its target question in the first sentence.

Cross-reference the output with SEMrush's Prompt Research tool, which groups related AI prompts by topic and lets you identify which prompts drive AI citations for your competitors versus your site.

The competitive AI audit — do this now.

Open ChatGPT, Perplexity, and Gemini. Type the 5 questions your most important customers would ask AI about your topic. Screenshot the results. Note:

• Which brands appear in the answers
• Which specific pages those brands are citing
• What language patterns the AI uses in its framing of your topic
• Whether your brand appears at all

This is your GEO competitive baseline. Brands that rank #1 on Google for a decade are being skipped because an AI cites a competitor instead. That gap is your content strategy. The pages getting cited are showing you exactly what to build.

SEMrush's AI Visibility Toolkit automates this audit at scale — tracking your brand's citation rate across ChatGPT, Gemini, Perplexity, and Google AI Overviews, monitoring sentiment, and surfacing the specific prompts where competitors appear and you don't. For agencies managing multiple client sites, this is worth the add-on.

Third-party presence — the layer most practitioners skip.

AI engines don't only learn from your website. They learn from everywhere your brand appears. NP Digital's research and GEO frameworks consistently find that AI citation rates are driven by a combination of on-site content quality and off-site brand presence: LinkedIn content, YouTube, Reddit threads where your topic is discussed, mentions in industry publications. An AI engine summarizing "best digital marketing practices for MENA" is pulling from dozens of sources — not one. If your brand only exists on your own domain, you're one of those sources competing with specialists who live on ten platforms simultaneously.

For new sites specifically: the GEO timeline is 8–12 weeks for citations to stabilize. You cannot compress this with technical tricks. You can accelerate it by earning mentions in credible publications, maintaining consistent LinkedIn activity, and building the kind of third-party footprint that AI engines treat as authority signals.

The llms.txt question — honest answer

You've seen llms.txt in your feed. Every GEO guide mentions it.

Here's the honest 2026 read from the best available evidence: it's not a ranking factor yet. A study of the 50 most AI-cited domains found only one of them had an /llms.txt file. Major LLM providers haven't issued formal guidance. The overlap with actual AI citation behavior is currently near-zero.

But NP Digital's Senior SEO Manager Anna Holmquist frames it well: llms.txt files don't shift rankings today, but they set early rules for how your content interacts with AI systems — think of it like robots.txt in its early years: small now, foundational later.

Create one. Place it at yourdomain.com/llms.txt. List your most important pages with one-line descriptions of what each covers. Half a day of work. No downside. Over 844,000 domains have already adopted the format. When standards crystallize — and they will — you'll be already-correct.

What this reveals about the bigger shift

Twenty-five years of SEO was a game about position. Get to the top of the list. Own the keyword. Drive the click.

GEO is a game about trust. An AI engine is deciding — across thousands of possible sources — which handful to name in its answer. It doesn't produce a list. It produces a recommendation. And recommendations are how trust is expressed.

The implication: the gap between ranking and being cited will widen every year as AI adoption grows. The sites that invest in citability now build a compounding structural advantage. Gartner's projected 25% traditional search volume decline doesn't happen evenly — it happens fast in the query categories where AI answers are strongest (informational, research-heavy, comparison-driven). Those are exactly the categories that have always driven top-of-funnel traffic. The sites that own those categories in AI will own discovery before users ever form an intent to click.

Citation authority, like domain authority before it, takes time to build. The window to establish first-mover advantage in most niches is still open. Nearly 47% of brands have no GEO strategy. That won't be true in 18 months.

The complete GEO launch checklist

For new sites and for sites doing a technical GEO review:

Week 1 — Access

• Check robots.txt for AI crawler user agents (GPTBot, ClaudeBot, Google-Extended, PerplexityBot)
• Audit Cloudflare Bot Fight Mode settings if applicable
• Verify core content pages are server-side rendered (check source HTML for body copy)
• Submit sitemap to Google Search Console
• Set up GA4 AI referral source filters (create custom channel for referrals from chat.openai.com, perplexity.ai, gemini.google.com, claude.ai)
• Create /llms.txt with priority pages listed

Week 2 — Structure

• Deploy FAQPage schema on question-answering pages
• Deploy Article schema on all long-form content with author, datePublished, dateModified
• Rewrite H2/H3 headings as direct questions
• Rewrite opening paragraphs of top 5 pages to answer-first structure
• Add at least one cited statistic to every content section
• Validate all schema at Rich Results Test and Schema.org validator

Week 3 — Content architecture

• Run AnswerThePublic for your core 3 topics — export and map question clusters
• Run competitive AI audit: 5 key prompts across ChatGPT, Perplexity, Gemini
• Identify top 3 content gaps where competitors are cited and you're not
• Create or update content for those 3 gaps with answer-first structure and specific statistics
• Ensure consistent brand name and entity clarity across LinkedIn, Google Business Profile, and any directory listings

Ongoing — Track

• Run the 5-prompt competitive audit monthly
• Use SEMrush AI Visibility Toolkit or equivalent for citation tracking
• Update dateModified in schema whenever content is refreshed
• Treat content as a living asset — quarterly reviews minimum

The close

nvm.marketing goes live this week.

The second checklist is running alongside the first one right now. Cloudflare settings — verified. AI crawlers — unblocked. llms.txt — drafted and ready to publish at nvm.marketing/llms.txt on launch day. Schema — deployed on the first articles before the domain goes live. The opening paragraphs on key pages are already written to the answer-first structure.

The citation timeline is 8–12 weeks from launch. I'll report back at Issue #8.

In the meantime — run your own audit. The 20-minute version is free. Open ChatGPT. Type the question your ideal client would ask. See who appears.

If it's not you, you just found your content strategy.

PROMPT OF THE WEEK

Build your GEO prompt map before you write a single page.

"I run a website about [your topic/niche] targeting [your audience]. List the 20 most specific questions this audience would ask a generative AI assistant — not a search engine — about [your topic]. Write them as exact conversational prompts, not keyword phrases. Then group them into 5 content themes. For each theme, identify the single most important question an AI would use as the 'primary sub-query' when synthesizing an answer. This is my GEO content architecture."

What this gives you: a complete content structure built around AI retrieval patterns rather than keyword volume. Cross-reference with AnswerThePublic for real autocomplete data on each cluster. Your first 5 pages should directly answer the 5 primary sub-queries. Everything else supports them.

THIS WEEK

AI-referred sessions up 527% year-over-year — but only to sites built for retrieval. Previsible's 2025 AI Traffic Report tracked 19 GA4 properties. LLM-referred sessions grew from ~17,000 to 107,000 in a year. Conversion quality from AI referrals significantly outperforms traditional organic. The traffic is there. The question is whether your site is structured to capture it — or whether AI engines are routing users around you to someone whose content is more extractable.

Overlap between top Google rankings and AI-cited sources has collapsed from 70% to under 20% — 5WPR, "GEO vs. SEO: The 2026 Venn Diagram," 4 May 2026, based on more than 680 million citations collected between August 2024 and April 2026 across the five major AI platforms.

Only 44.3% of Google top-10 pages appear in AI answers — Semrush February 2026 study. Semrush analyzed 10 SaaS query sets across 4 AI platforms in February 2026. Fewer than half of pages ranking in Google's top 10 showed up in any AI-generated response. The gap is structural, not cyclical — it reflects different evaluation criteria between ranking algorithms and retrieval systems. Ranking first doesn't mean being cited. These are now different competitions.

Google Marketing Live — May 20, 9 days away. Six significant Google Ads announcements in 5 days before the keynote: journey-aware bidding in beta, Smart Bidding Exploration to Shopping and Performance Max, demand-led pacing, Data Manager Map View, Meridian Studio, Meridian GeoX. The pre-GML cadence is the densest ever. The keynote on May 20 (8:45 AM PT) will build on all of it. EMEA session follows May 21. Register free. If the GML announcements land the way the pre-announcements signal, Issue #5 will cover the full picture.

AnswerThePublic AI Dashboard shows how AI currently describes your keyword — not how people search it. Neil Patel's question research platform added AI Dashboard functionality in 2026. It now shows the framing AI systems use around any topic — not just the search patterns humans produce. For GEO content planning, this is the distinction that matters: knowing how an AI describes your category before you write content about it. Free tier available. Integrated with Ubersuggest.

TOOL OF THE WEEK

AnswerThePublic — The prompt map that should exist before your first page

If you do one thing from this Deep Dive before next Monday, make it this.

What it does: AnswerThePublic pulls autocomplete data from Google, YouTube, and Bing and maps it into visual question clusters — who, what, when, where, why, how, which, versus, like. Every spoke is a real question your audience is typing. The new AI Dashboard (2026) adds a layer showing how AI systems currently frame your topic, so you can see the language patterns AI uses before you write content meant to appear inside AI answers.

The GEO-specific workflow: Enter your core topic. Pull the full question export. Filter for question-format queries specifically — the "how," "what," "why," "when" prompts. These are your retrieval targets. AI engines fan out user questions into exactly these sub-queries during retrieval. If you have content that directly answers them, with a cited stat and an answer-first opening sentence, you get retrieved. If you don't, whoever does gets cited in your place.

Why it matters for a new site: The difference between a site that gets AI citations in 10 weeks versus 6 months is mostly content architecture. Sites that launch with content built around real question clusters and answer-first structure earn citations faster because they're already structured for retrieval. AnswerThePublic gives you that architecture before you write. Build the map first. Write to the map.

Cost: Free tier is sufficient for initial topic research. Full export and AI Dashboard functionality requires an Ubersuggest subscription.

MY TAKE

nvm.marketing goes live this week. The second checklist is running right now, before the domain is public.

That sequencing matters. Most practitioners run the GEO audit after launch, if they run it at all. The access layer, the schema, the answer-first structure — these are things people retrofit months later when they notice AI search isn't sending any traffic. I'm running them before the first page is indexed.

That's the point of this Deep Dive. Not theory. Not retrospective. A live checklist, documented in real time, for a real site launching this week.

The part nobody tells you about GEO is that the work is mostly unglamorous. It's robots.txt hygiene. It's schema deployment. It's rewriting your first paragraph so it answers the question before it builds to the answer. None of it is technically difficult. Most of it is just not done.

47% of brands have no GEO strategy. That's not because GEO is new and complicated. It's because nobody sent them the checklist.

Now you have it. I'm running it this week. You should too.

That's Issue #4 of NVM — Deep Dive Series I, Vol. 01. Every 4th issue, we go deep on one topic. nvm.marketing launches this week — come find your seat. The field guide format returns at Issue #8. The regular issue is back next Monday. If this gave you one thing to act on this week, share it with someone who needs it.

Reply to this if you have a question — I read every one. See you next week. — Mostafa

NVM · Front row seats to AI and digital marketing