Every week the timeline claims “SEO is dead, GEO is the new gold rush.” While preparing to launch @SpaceCutAI (an AI video editor), Theo (@TheoAGI) wanted engineering truth rather than hype. He had Claude Fable 5.2 read, cross-check, and sort 66 bookmarked SEO/GEO articles, 48 free AI agent SEO skills, and 14 GitHub repositories (2,000–27,000+ stars), then ran the checks directly against his pre-launch site. Here is the complete synthesis—plus the exact audit prompt at the end to hand straight to your AI.
4 Things Nobody Tells You
About 90% of the advice repeats the same dozen ideas
Strip away the rebranded buzzwords and roughly 90% of SEO/GEO threads recycle the exact same dozen ideas—most of which are classic SEO.
Almost none of the headline numbers come with real data
In at least two viral articles, the "AI mentions" touted in the headline turned out to be plain Google Search Console impressions—admitted by the article itself further down.
Quoted thresholds are rules of thumb, not hard laws
Thresholds like "position 8 to 15", "5 internal links", and "DR 20" are rules of thumb. Honest practitioners state that clearly; hype accounts present them as hard facts.
GitHub stars don't tell you if the advice is current
One repository with over 11,000 stars gets ChatGPT's live search crawler completely wrong. Popularity does not equal technical currency.
1.The Foundation: Make Sure Google & AI Engines Can Read You
Before worrying about content velocity or backlinks, traditional search crawlers and live AI retrieval bots must be able to fetch and parse your pages cleanly:
Inspect raw HTML page source, not just DevTools rendered DOM
Open your page source. If your headline and main body text aren't present in the raw HTML, any AI crawler that skips JavaScript execution sees an empty page.
Separate model-training crawlers from live-answer search crawlers in robots.txt
GPTBot, ClaudeBot, and Google-Extended collect training data. OAI-SearchBot, Claude-SearchBot, PerplexityBot, and Bingbot power live answers. Block the second group and you vanish from AI search.
Read your live deployed robots.txt—never trust the file in your repo alone
Cloudflare can prepend its own managed rule block at the top of your live robots.txt and block AI bots with a single dashboard toggle. Always fetch the live URL.
Ship clean sitemaps, canonicals, and honest Schema—zero fake review stars
Add a sitemap containing only indexable pages, a canonical tag on every page, and structured data (Organization, WebSite, plus Product or SoftwareApplication). No review stars until you have real reviews.
| Crawler Group | User-Agent Tokens | Core Function | Impact of Blocking |
|---|---|---|---|
| Model Training Bots | GPTBot, ClaudeBot, Google-Extended | Collects corpus data for future model training weights | Optional policy choice; does not kill live search retrieval |
| Live Answer Search Bots | OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot | Powers live RAG answers and clickable source citations | NEVER BLOCK! Anything other than HTTP 200 erases you from AI search |
curl -A "OAI-SearchBot" -s -o /dev/null -w "%{http_code}" https://yoursite.com/Theo’s pre-launch confession: "I had Claude run these checks on my own site before launch. Cloudflare had added nothing to my robots.txt. But I had no sitemap, no canonical tag, no structured data, and my www address didn’t redirect to the main one. I thought the site was ready."
2.Bing Is Back: Free First-Party AI Citation Data in 10 Minutes
Set up Bing Webmaster Tools today: import directly from Google Search Console, submit your sitemap, and turn on IndexNow. Total setup time: ten minutes.
Copilot runs directly on Bing, and ChatGPT has leaned heavily on Bing infrastructure for live search.
Pushes newly published or updated decision pages straight to participating search engines.
Shows the exact queries where AI answers cited your site—the closest thing available to free, first-party citation data.
3.Build Pages for Decisions, Not for Keywords
The pages that consistently earn AI citations—and actually convert into revenue—are the ones buyers read right before choosing:
The 7 Quantitative Decision-Page Rules (Distilled from 48 Skills)
- 6 / 10Google the query first. If 6 or more of the top 10 are the same type of page, that's the type to build. If they're all blog posts and you sell a product, skip it.
- 4 / 10Two keywords belong on the exact same page when they share 4 or more of their top 10 Google results.
- 40–80wAnswer the query directly in the first 40 to 80 words and name your product. AI engines lift passages, not whole pages.
- 5–8A "best of" list has 5 to 8 options. 3 reads like an advert; 25 becomes a directory.
- DatedEvery price comes from the vendor’s own page, marked "checked on" with a date. Include one real drawback per option—yours included.
- No #1Don't rank yourself first in your own list.
- +1 GainAdd one thing the top 10 results don't have: your own number, your own test, your own screenshot—and show the method.
Interactive Sandbox: SERP 6/10 Page-Type & 4/10 Clustering Rule
4.GEO: What's Actually New Beyond Classic SEO
Assistants run ~10 hidden searches per question
Expand the "searching" step inside ChatGPT or Claude and read the sub-queries directly. Every single hidden query is either a page you need to build or a third-party site you need to be listed on.
85% third-party citations vs. <4% brand sites
In a dataset of 44,070 citations across 10 brands, 85% went to third-party pages ('best of' lists, YouTube, review sites) and under 4% to the brand's own site. Yet brands whose own site got cited were named far more often. You need both.
Answers move weekly—a single test proves nothing
Studies quoted across these articles show ChatGPT swaps most of its cited sources from one week to the next. Never celebrate or panic over a single prompt run.
Dated Brand Fact Sheet & the Two-Month Fix Rule
AI frequently gets your pricing or platforms wrong. Publish a dated fact sheet, trace each wrong claim to the external page it came from, fix it there, and call it fixed only when gone two months in a row.
5.Measure It for Free: Skip Paid Visibility Trackers Early On
5-Step Manual Attribution Loop
- 1.Write 15–30 real buyer questions pulled from your inbox, sales calls, and reviews—never your imagination.
- 2.Run each question at least twice per engine while logged out: ChatGPT, Claude, Perplexity, and Gemini.
- 3.Log both who got named and which URLs got cited. Being cited and being recommended are two different numbers.
- 4.Under ~100 AI visits a month, report raw counts, never percentages.
- 5.Add "How did you hear about us?" at signup. One founder found analytics credited Google with 3.4x more signups than customers actually reported.
The Complete Zero-Cost Tool Stack
6.Links and Mentions: Worth Your Time vs. What Gets You Burned
- 1. 20–30 Launch Platforms & Software Directories: Use the exact same name, one-line description, and logo everywhere. Do 20 to 30—not 300.
- 2. Unlinked Brand Mentions: Search
"your brand" -site:yoursite.comand ask for the link. - 3. Competitor "Best-Of" Lists Missing You: Send a concise 120–180 word pitch backed by evidence. Follow up once after a week, then stop.
- 4. Vendor Customer Stories: Write a customer story about a tool in your stack—vendors love linking back to real case studies.
- 5. Disclosed Founder Replies on Reddit & Quora: Post as yourself, fully disclosed. At most one answer mentioning your company per subreddit per week.
- 1. Link Exchanges & Credit Networks: Even the articles selling automated link-exchange networks admit Google classifies them as link spam.
- 2. Paid Reddit/Quora Tasks, Fake Accounts & Self-Written Reviews: Triggers permanent subreddit bans, domain blacklisting, and trust destruction.
- 3. 1,500 Scaled AI Posts Across Site Networks: Squarely targets Google’s Scaled Content Abuse site-wide demotion systems.
- 4. Doorway "City × Service" Template Pages: Classic doorway-page footprint that burns crawl budget and tanks site-wide quality scores.
7.If an AI Agent Does Your SEO: Permission Tiers, Hooks & 32/40 QA
Numbers must come from the tool, never model memory. Test it by asking for your top 5 pages by clicks and checking GSC.
Grade every draft before shipping: 8 areas out of 5, passing at ≥32/40 with no single area under 3.
Enforce "never" in a hard code hook, not a prompt. Always inspect the diff and snapshot key pages before/after deploy against the 4 silent killers.
8.New Domain? Slow Down — Plus the Exact Execution Order
• 1 to 4 new pages a week: Practitioners disagree slightly on the exact count, never on the direction.
• Wait 2 to 3 months before judging a change: Give crawlers and ranking signals time to settle.
• Save GSC optimization playbooks for Month 3: Striking-distance tweaks, title tests, and content pruning require Search Console history a brand-new domain does not have yet.
Complete Section 1 foundation, set up Bing + IndexNow, and run your first baseline of buyer questions.
Verify Search Console, submit your sitemap, and ship consistent listings across 20–30 directories.
Publish one high-intent decision page per week, paired with one chaptered, hand-captioned video per page.
Open Search Console with real impression history and begin systematic striking-distance optimization.
The Two Tools Behind 49 of These Articles (ChatSEO vs. Distribb)
Full disclosure from Theo: 49 of the 66 articles were written by the founders behind Distribb (@borjafat) and ChatSEO (@NicholasDulait & @VengeonsP). Everything in this playbook can be done by hand; these tools automate the three bottlenecks that are slow or costly manually (as read on Nov 10, 2026):
| Dimension | ChatSEO (@NicholasDulait / @VengeonsP) | Distribb (@borjafat) |
|---|---|---|
| Core Strength | Pulls live volumes, backlinks & competitor data via DataForSEO next to read-only GSC + site crawl prioritized fix list | Tracks daily when ChatGPT, Gemini, Perplexity & Google AI answers cite your site |
| CMS Integrations | WordPress, Shopify, Webflow | WordPress, Shopify, Webflow, Wix, Ghost, Notion, Framer |
| MCP & Pricing | Starts at €29/mo (€49/mo Pro with MCP server), 7-day trial | $97/mo (MCP server included on every plan), 3-day trial |
The 14 GitHub Repos, Sorted & Fact-Checked
Quick litmus test for any SEO skill: ask which bot controls ChatGPT Search visibility. OpenAI’s docs say OAI-SearchBot (GPTBot is for training). Also, 3 repos output a score out of 100 whose weights were never validated against real rankings—read the findings, ignore the score.
The clearest method of the entire batch, MIT-licensed, with nothing sold inside the skill files.
Most precise on technical checks, structured data & AI crawlers with real scripts. Gets OAI-SearchBot right and enforces "Score only what was measured." Use page, technical, and schema commands.
Connects Claude to your Search Console. Read-only by default, runs locally on your machine, free.
Open-source alternative to Semrush/Ahrefs. Best line: "Search volume is not visits." Requires paid keyword data.
Solid single-page audit without GSC data, but routes through their hosted server and still rates FAQ markup as "High Impact" years after Google restricted FAQ rich results.
Keep ONLY its live Cloudflare robots.txt check. Its crawler doc wrongly claims blocking GPTBot removes you from ChatGPT Search, lists dead 2023 HowTo markup, and pipes curl straight into bash.
Keep the brand-voice/features context file layout. Drop the "1–2% keyword density" target and delete the line asking for named mini-stories "even if fictional".
One great rule—"Never invent volumes"—but thin skills and data flows through their hosted connector.
Clean vendor-free reading list. Best tip: check Cloudflare Security Events after launch so Bot Fight Mode isn’t challenging legitimate crawlers.
The 10 Skills I'd Start With (All from RomeoSEO/seo-skills)
Skip the remaining 38 skills (local stores, migrations, or multi-month GSC workflows) until you actually need them. Start with these 10:
Full crawlability, raw HTML rendering, redirect chains, and indexation blocker audit
Clean Organization, WebSite, and Product/SoftwareApplication JSON-LD without deprecated bloat
Verifies live search bot access (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot)
Ensures XML sitemaps contain strictly 200-OK canonical indexable URLs
Pre/post-deploy snapshot diffing against the 4 silent site killers
Builds high-converting decision pages with 40–80 word opening answers and verified pricing
Extracts 15–30 real buyer questions and maps hidden assistant fan-out sub-searches
Maintains a dated brand fact sheet and traces hallucinated claims back to source URLs
8-area × 5-point rubric gatekeeper requiring ≥32/40 with no area below 3 before shipping
Combines Bing AI Performance, GSC, and signup self-attribution to measure true AI ROI
Hand This to Your AI: Interactive 7-Step Zero-Hallucination Audit Prompt
You don't have to run every check manually. Fill in the three lines below, copy the prompt, and paste this article underneath it into an agent with repo access (Claude Code, Cursor, Codex)—or use a chat assistant to run the read-only audit and draft outlines:
Apply the playbook below to my site. Site: https://yoursite.com Product: An AI-powered video editing tool that removes silences and adds captions automatically Competitors: Descript, CapCut, OpusClip Work in this order. Stop after each step and show me what you found. 1. Audit only. Change nothing. Fetch my live robots.txt, the raw HTML of my homepage, my sitemap and my www address. For every point in section 1, say pass or fail, with the proof: the line or the status code. 2. List the fixes, smallest first. For each one: the file, the change, and how you'll check it worked. 3. Wait for my yes before each change. Show me the diff first for anything that touches robots.txt, canonicals, redirects or noindex. Never delete a page. 4. After each change, run the check again and show me the result. 5. Give me the list of what only I can do (Search Console, Bing Webmaster Tools, directories, outreach, videos), in order. 6. Draft 20 buyer questions for my product. Mark the ones you guessed, so I can replace them with real ones. 7. Propose 5 decision pages from section 3. For each: the query, what the top 10 results look like today, and the one thing my page would add. Don't write any page until I pick one. Rules: - A number comes from a tool or a page you fetched, never from memory. If you can't check something, write "not checked". - Never invent a search volume. - No review stars, no FAQ markup, no HowTo markup. - Install no skill or repo from this article without showing me the file first.
Sources and Practitioner Credits
As Theo notes: "Selling a tool doesn’t make someone wrong, and an outdated line doesn’t make a repo useless." Every actionable rule in this playbook comes from practitioners who shared their work publicly: