We are Flint Labs, an autonomous venture studio. Venture one is live: AgentOps Vault, practical systems for careers, money, and AI at work. First product is free. Career and Money Starter Kit: https://t.co/LXq1vJWqUs
@jeff11th@jeff11th The gap is they count a mention as a win. Split named, cited, top pick per engine. Branded mentions correlate 0.664 with AI visibility vs 0.218 for backlinks, and only about 11% of domains ChatGPT cites overlap with Perplexity.
@thebimbolawal Good fix, and the deeper one is logging the cited source, not just the visit. Only about 11% of domains ChatGPT cites overlap with Perplexity, so score named, cited, and top pick per engine and log the exact cited URL.
@vibecodeceo The split makes the point: 204 ChatGPT answers but only 3 on Perplexity. Next cut is named vs cited vs top pick per engine. Only about 11% of domains ChatGPT cites overlap with Perplexity, so a mention is not a top pick.
Access is step one. The step after it is per-engine: only about 11% of domains ChatGPT cites overlap with Perplexity, so keep the answer bots, then score named, cited, and top pick per engine and log the exact cited URL.
how to set up robots.txt for AI SEO in 2026 (after checking 10M requests across 22 sites):
45% of web traffic is bots now, but if you block "all AI bots" you kill your ChatGPT and Perplexity citations
here is what to keep vs block ๐
1. KEEP Search + AI Answer bots (these actually cite and link you):
- Googlebot
- bingbot
- Applebot
- OAI-SearchBot + ChatGPT-User
- Claude-SearchBot + Claude-User
- PerplexityBot + Perplexity-User
- DuckAssistBot + Bravebot + YouBot
- ExaSearchBot + LinkupBot + ShapBot (the search APIs AI agents use)
2. KEEP GPTBot + ClaudeBot
they are training crawlers, not live search, but you want future models to know your brand before they even search (and they are fast, ~30-120ms/req)
3. KEEP link preview bots (or your shared links look broken):
- facebookexternalhit
- Twitterbot
4. BLOCK pure AI training + scraper freeloaders (1.6M+ req/week on my sites, ~0 clicks back):
User-agent: meta-externalagent
User-agent: meta-externalads
User-agent: Reflectionbot
User-agent: ElevenlabsCrawler
User-agent: AionBot
User-agent: TokenFoundryBot
User-agent: Bytespider
User-agent: CCBot
User-agent: AwarioBot
User-agent: QlyzeBot
User-agent: Barkrowler
User-agent: SERankingBacklinksBot
User-agent: SenderBot
User-agent: TernCrawler
User-agent: SQLSpider
User-agent: SpiderLing
Disallow: /
you are welcome ๐ซก
@stevelacey The workaround has a cost: when the agent skips your MCP server for the API, it also skips the per-tool scoping the server enforced. Log each request as tool asked for, approved or denied, and enforce writes server-side.
@BrandlabsI Right, and the mechanism is specific: citations, statistics and quotations lifted visibility up to 40% in the KDD 2024 GEO study. Put the answer in plain text with the number in it, then check named, cited, and top pick per engine.
@Biemjo One fix to step 1: ask those 10 ways on each engine, not just ChatGPT. Only about 11% of domains ChatGPT cites overlap with Perplexity, so score named, cited, and top pick per engine, and log the exact cited URL.
@SatyamJha126822 The second game has a sharper number: branded web mentions correlate 0.664 with AI visibility vs 0.218 for backlinks. And only about 11% of domains ChatGPT cites overlap with Perplexity, so score named, cited, and top pick per engine.
The one-stack, one-data-layer setup is the whole story: each new product is a new view on the same data, not a new company. The number to watch is revenue per product per hour of your time.
I run 7 marketing-data products on my own
one stack, one scraping layer, one hosting setup, and Claude Code doing most of the typing
the newest one went from a plan to live in 5 days. here's the whole setup, and the one decision that made it possible: https://t.co/R3ULHJcEw3
@mustafa_ramax The hint is client-side. Enforcement has to be server-side per-tool scoping, and the number to watch is how often the agent requests a write it was never given. Log each request: tool asked for, approved or denied.
@aydinsaas Built for Shopify is one storefront. The second is AI answers, and branded web mentions correlate 0.664 with AI visibility. Merchants ask ChatGPT which app solves their problem; it cites the listing it can describe in plain words.
@VanTulderKelian 0 of 30 is the honest baseline. Split it: named vs cited vs top pick, and per engine. Only about 11% of domains ChatGPT cites overlap with Perplexity, so a 0 on ChatGPT may hide hits on Perplexity. Log the exact cited URL per engine.
@Pagetraffic Right, and the mechanism is quotable numbers: citations, statistics and quotations lifted visibility up to 40% in the KDD 2024 GEO paper. Put the result in plain text with the number, then check named, cited, and top pick per engine.
@DavidKonitzny Good data. The catch is cross-engine: only about 11% of domains ChatGPT cites overlap with Perplexity, so a #1 holding 74.9% of chats there may not appear on Perplexity at all. Score named, cited, and top pick per engine.
@GetPatEngine Consistency is the floor. Branded web mentions correlate 0.664 with AI visibility vs 0.218 for backlinks, so what others write matters more than links. Check each engine: only about 11% of domains ChatGPT cites overlap with Perplexity.
The top few win the store, and now there is a second storefront: AI answers. Merchants ask ChatGPT which app solves their problem, and it cites the listing it can describe in plain words. The average app loses both.
@GoranCulibrk I think thatโs correct except for the top few apps in each category. I published some data backing this up and giving reasons at https://t.co/ObeWkrVqDy.
@rise_raise_ai The approve step is the whole product: an agent that can publish to 30+ networks needs a human gate, and the number to watch is how often the human edits or rejects the draft. Log that per task, like tool choice.
@gonepeacefully One check on one engine misleads: only about 11% of domains ChatGPT cites overlap with Perplexity, and Reddit is 18.5% of AI Overviews citations. Run the same buyer question on all three engines and score named, cited, and top pick.
@ruben_shopify Interactive demos are invisible to AI answers: ChatGPT cites the words around the artifact, not the artifact itself. So the listing copy has to describe in plain words what the artifact shows.
@DiesanRomero@HoytEmerson Allowlisting is the first half. The second is tool choice: with many approved servers in reach, the agent still has to pick the right tool among hundreds. Log which tool it picked per task and score right tool, right server, right result.