Gensight.AI (@Gensight_AI) header
Gensight.AI (@Gensight_AI) profile picture

Engineering your AI visibility. The world's first GEO platform designed for both brands and influencers.

Manchester, England
Joined March 2026
AI agents lean on a few trusted sources. Now some big publishers sell brands space in a version of the page only AI reads. In one study, agents picked the worse option 68% of the time when it came from a source they favoured. For now, brands can buy their way in.
1
9
Our own logs, past 36 days: AI crawler visits: 1,687 robots.txt: 346 sitemaps: 96 llms.txt: 0 It became a GEO checklist staple, ours included, without evidence AI systems read it. We'll be reducing its weight in our score.
1
7
How ChatGPT decides which brands to name: 1. Turns your question into filters 2. Decides whether to search the web 3. Uses what it learned in training, plus what it can read if it searched 4. Keeps what several sources agree on 5. Recommends, mentions or cites
1
5
We ran one of the best known food travel creators in the world through our influencer-mode audit. 57/100, sixth among his peers, behind creators with much smaller audiences. The engines credit him on 2 of the 5 topics that decide credibility in food travel. One he never gets credited for: "engaging storytelling & personality" - the thing he's best known for. The three who got that credit are all writers. Their storytelling is written down. His is a facial expression in a video. Being famous and being quotable are two different things.
14
On Monday, Cloudflare (in front of ~a quarter of all websites) changes its AI crawler defaults. One "block AI bots" switch becomes three: Search, Agent, Training. On ad-carrying pages, Training and Agent are blocked by default. And mixed-purpose crawlers are judged by their strictest behaviour, i.e. blocking Training also blocks Googlebot. Behind it: 52% of crawler requests were for AI training by June, up from 22% in spring 2025. Anthropic's crawl-to-referral ratio sits above 20,000 pages fetched per visitor sent back. So what happens to AI answers when the open web closes to crawlers while machine-written pages keep climbing? Find out more on the Gensight.ai knowledge-hub
11
AI Traffic analytics is live on GenSight.AI. Audits measure what AI can read on your site. This is the record of what it did read - observed at the edge, split into training, live retrieval, search indexing and agent connections. Coverage names the pages no AI has fetched in 30 days, per company. Useful two ways: make those pages retrievable, or move the messaging onto pages engines already read. Being read is the first step. From there it's being understood, then cited, then recommended, and that's the path we build the roadmap around. Also: it will finally show whether anything requests llms.txt. Unproven standard, now measurable.
2
10
From our co-founder, Tamanna Haque: "Excited to reveal our first time publicly speaking about GenSight.AI, on Nune Isabekyan's Root Cause Podcast! We talked about what happens when people stop Googling and start discovering brands, products and people through AI. We got into what AI Visibility actually is; Nune came in openly sceptical, which made it a much better conversation. We covered whether GEO just feeds the AI slop machine, how you get a deterministic score out of non-deterministic models (inspired by credit scores), practical GEO advice for content creators, why working on AI Visibility should be an ongoing priority, and more. Thanks Nune for having me, and for the challenge." Listen here: thoughtfultechnologist.com/p… or directly on Spotify here: open.spotify.com/show/033eSy…
29
Our first case study is live. GGK Technologies, an early-stage incident-reporting SaaS, worked through seven roadmap items in-house with no agency and no ad spend. Baseline 34 in April, 55 by August. Retrieval surface 60 to 90, Entity Strength 10 to 45. Same engine, same 35 signals, same weights every run - the score moved because the inputs did. Are they ChatGPT's top answer yet? Not yet, and no honest tool would claim it. Read you, know you, recommend you: they've built the first two, and the third compounds from here.
11
Ran our new per-product audit on a famous brand's flagship coffee machine. Comprehension: 84/100. AI knows exactly what it is. Competitive position: 19. Mentioned in 1 of 6 category queries - only when asked by name. Understood perfectly. Recommended to nobody. Brand visibility ≠ product visibility.
9
Nobody buys from one prompt. Real sessions are arguments - budget added, pushback, "what's the downside?" Getting named in turn one is easy. A lucky citation does it. Then the follow-ups start, and a brand the model doesn't deeply know gets hedged: "some users report..." A hedge loses to conviction. Every time. AI seeing you is not AI believing you. And purchases live at the end of conversations.
8
Loyalty is about to become a setting. Your customer's agent re-evaluates every repurchase at zero effort. But people will set defaults - "my usual, unless the price moves." So every category splits in two: brands named in the config, holding a default that defends itself silently every cycle. And everyone else in the auction - re-specified from scratch, every purchase, on machine-readable criteria. "Do customers love us?" is the old question. The new one: are we in the config, or in the auction?
4
AI slop didn't fool the engines. It taught them who not to trust. When text became free to generate, self-published content stopped counting as evidence. Engines discounted the channel - trust moved to third-party sources and verifiable facts. The proof: when AI Overviews cite a brand's own "best of" listicle, they recommend a competitor 69% of the time. Your page briefs the engine; the verdict comes from elsewhere. And it doesn't revert. Text isn't going to stop being free.
4