GEO papers for your clients
Six short papers that explain GEO — how AI assistants read websites — in plain English, branded with your agency name, address, and logo. Pick a paper, then use Download PDF to save or print just that one. Send them to prospects or bring them to meetings.
What Is GEO?
GEO stands for Generative Engine Optimization. It means making your website easy for AI assistants — ChatGPT, Gemini, Perplexity — to read and understand. That reading is the entry ticket: an assistant can't name a business it couldn't learn about — though being readable is a start, not a guarantee. This paper explains, in plain English, how those assistants decide which businesses to name.
Search changed. The answer is now the result.
For twenty years, customers typed a question into Google and picked from a list of links. More and more, they now ask an AI assistant — and the assistant answers directly, naming one or two businesses it trusts. There is no page two. If the assistant can't read your website, you are not "ranked lower." You are simply not in the answer.
The good news: the assistant isn't guessing. It builds its answer from your website — your services, hours, prices, reviews, and answers to common questions. GEO is the work of making sure all of that is on your site, as plain text a machine can read.
Most websites aren't ready
We scored 10,369 real small business websites on how well AI can read them (0–100). Here's what we found:
Out of every 10 small business websites, 5 score below 70. The average score is 67.1 out of 100, and the average site has about 12 separate issues holding it back (an issue is one failed check the scanner records; the exact average is 11.7). Almost none of them are hard to fix — most owners just don't know they exist, because the site looks fine to a human.
Source: CrawlReady AI-readiness study, October 2026 — 13,342 small business websites attempted; 10,369 loaded and were scored with the v2 scorer. How the score works: five checks total 90 points — crawler access 20, structure 25, metadata 25, content 15, performance 5. An FAQ adjustment makes up the rest: a FAQ score above 70 adds 10 points; 40 or below subtracts 10. A site that blocks every AI crawler is capped at 40, however good the rest is. Scores reflect what AI systems can read from a website, not traditional search rankings — and readability alone does not guarantee a recommendation.
The 5 Things AI Checks on Your Site
When an AI assistant reads your website, it effectively grades five things. Here they are in plain English — with the average grade from our study of 10,369 small business websites, so you can see where most businesses stand.
…and one bonus check almost everyone fails
AI assistants love question-and-answer content: it's exactly the shape of what people ask them. Yet in our study, 81 out of every 100 websites score 40 or below on FAQ signals — no real Q&A content an assistant can lift into an answer.
The pattern across all five checks is the same: the basics are usually in place, but the specifics AI needs — organized headings, labelled business facts, real answers to real questions — are missing. That's fixable, and it's where the advantage is: your competitors mostly haven't fixed it either.
Source: CrawlReady AI-readiness study, October 2026 — 10,369 small business websites scored with the v2 scorer. Category averages: crawler access 98.1, performance 99.0, content 84.2, metadata 80.8, structure 70.5; FAQ signal average 43.9, with 81.4% of sites at 40 or below.
Are You Blocking AI by Accident?
Your website has a door policy for robots — a small file called robots.txt, plus settings in your security tools. If it turns AI assistants away, they can't read your site at all — and an assistant can't recommend a business it knows nothing about.
Meet the visitors
| Visitor | Who it works for | What it does on your site |
|---|---|---|
| GPTBot | ChatGPT (OpenAI) | Reads your pages so ChatGPT can answer questions about your business and cite your site. |
| ClaudeBot | Claude (Anthropic) | Same job for Claude — gathers the facts behind its recommendations. |
| PerplexityBot | Perplexity | Fetches pages live while answering, and quotes them as sources. |
| Google-Extended | Gemini (Google) | Controls whether your content can feed Google's AI answers and training. |
| CCBot | Common Crawl | Builds the public web archive many AI systems learn from. |
None of these visitors can buy anything, spam you, or change your site. They only read. Blocking them doesn't hurt them — it removes *you* from their answers.
How common is it?
In our study of 10,369 websites, about 2 in every 100 had crawler-access problems serious enough to zero out their AI visibility. That sounds rare — until it happens to you, when the loss is total. And in almost every case we examined, nobody chose to block AI. It happened by accident:
• A security plugin or firewall set to "block bots" — it blocks the good ones too.
• A line copied into robots.txt years ago, aimed at scrapers, that names all robots.
• A website builder or hosting default nobody reviewed.
• A "under construction" or login wall left in front of public pages.
The 2-minute check
Open your browser and go to yourwebsite.com/robots.txt (use your real domain). If you see a line that says Disallow: / under User-agent: * — or under any of the bot names above — your door is closed to that assistant. The fix is usually a one-line change, but do it with your web person: the same file also does legitimate work telling robots to skip private pages like your admin login.
Source: CrawlReady AI-readiness study, October 2026 — 10,369 small business websites scored. Crawler access: 219 sites (2.1%) had at least one AI crawler turned away; 152 (1.5%) blocked every AI crawler we tested. Bot names and roles as published by OpenAI, Anthropic, Perplexity, Google, and Common Crawl.
The 10 Fixes That Move Your Score Most
From our study of 10,369 small business websites: the issues that show up again and again, ranked by how many sites have them. Each fix names what to do — and what to send your web person if you'd rather not touch it yourself.
structure problem
Percentages are the share of the 10,369 studied websites with that issue (llms.txt absence is an estimate — the file is new enough that almost no small business has one). Notice the shape of the list: the top fixes aren't technical wizardry. They're your phone number, your address, and answers to your customers' questions — written where a machine can read them.
Source: CrawlReady AI-readiness study, October 2026 — 10,369 small business websites scored with the v2 scorer. Shares: weak FAQ signals 81.4%, no phone as plain text 81.0%, no address as plain text 80.3%, structure below 70 41.9%, thin content (under 300 words) 35.0%, metadata below 70 26.6%, crawler-access problems 2.1%.
How Your Industry Scores
Average AI-readiness scores by industry, from our study of 10,369 small business websites. The grey line is the overall average (67.1). Every industry has room to move — and the business that moves first in its market is first in line for the AI recommendations its competitors are missing.
What the spread tells you
Nobody is running away with it. The best-scoring industry (healthcare, 70.3) is barely into the "Workable" zone, and the whole field sits within about 10 points. These are not technology gaps — they're detail gaps. Healthcare sites tend to publish complete, structured information (services, locations, hours) because their platforms demand it. Retail sites lean on product photos and PDFs, which AI can't read as text — and it shows.
Your real competition is local, not average. When a customer asks an assistant for a recommendation, it compares you to the handful of businesses near you — not to a national average. If your site is the one in your town that AI can fully read, you're the one it can actually consider, even in a low-scoring industry. Especially in a low-scoring industry.
The cheapest points are the same everywhere. Across every industry we studied, the biggest losses come from the same short list: no FAQ content (81% of all sites), contact facts missing as plain text (about 80%), and weak page structure (42%). Fix those three and most businesses jump a full grade.
Source: CrawlReady AI-readiness study, October 2026 — 10,369 scored websites; industry identified for 4,011 of them (the rest unclassified, average 66.6). Averages are means of the v2 AI-readiness score (0–100). Small samples (under ~120 sites) should be read as directional.
Building a New Website? The GEO Checklist
A new website is the cheapest moment to get AI-readiness right — most of it costs nothing if it's in the plan from day one, and real money if it's bolted on later. Hand this list to your web designer before work starts, and check each box before the final payment.
Three questions to ask before you sign
1. "Show me a site you built and its AI-readiness score." A designer who works in GEO will know what you mean.
2. "Will my phone, address, and prices be real text — or pictures of text?" Only one answer is acceptable.
3. "What's included for AI crawlers — robots.txt, schema, llms.txt?" robots.txt access and schema belong in the base price, not an upsell; llms.txt is a five-minute extra worth including while it's still an edge.
Statistics from the CrawlReady AI-readiness study, October 2026 — 10,369 small business websites scored with the v2 scorer.