i built a tool that checks whether AI models recommend your product. then i pointed it at my own site.

vaulto.sh scored 0 out of 100.

not "needs improvement". not "below average". zero. neither Claude nor Perplexity knows vaulto exists. when buyers ask AI for an ai visibility scanner, we are not in the answer.

this post is the full autopsy: the real scan data, the uncomfortable lesson about why technical optimization wasn't enough, and the exact plan we're running to climb out. if you're a founder wondering why AI never mentions your product, this is for you.

key takeaways

an ai visibility score measures whether AI models actually mention and recommend you, not whether your site is well built
our site passes 82% of technical GEO checks and still scores 0 on visibility. technicals and visibility are different problems
AI models learn about products from what other people write about them. if nobody writes about you, you don't exist
when we asked AI about "vaulto", it cited HashiCorp's Vault docs, a crypto wallet's Trustpilot page, and Instagram reels about physical safes. not us
the fix is external: directories, comparison content, community answers, and building in public. on-site files only open the door

what does a 0/100 ai visibility score mean?

it means that when AI models are asked about your product or your category, you never come up. we query the models live with the questions real buyers ask: what tools do you recommend for X, what is [product], how does it compare. then we score whether you're mentioned, how prominently, and whether the model would recommend you.

vaulto's result: Claude scored us 0. Perplexity, which searches the live web before answering, also scored us 0. aggregate: 0/100.

zero isn't unusual, by the way. most small products score it. AI assistants answer from training data and from what they find when they search, and both of those depend on a footprint most early products haven't built yet. that's exactly why the score matters: the products that fix this early get recommended while their competitors stay invisible.

the embarrassing part: our technicals are nearly perfect

here's where it gets interesting. the same scan audits your site against the technical GEO checklist, the stuff every guide tells you to do. vaulto scored 82/100:

robots.txt with AI crawlers allowed: 18/18
meta tags: 14/14
structured data (JSON-LD): 14/16
llms.txt: 13/18
content signals: 9/12

we did the homework. GPTBot, ClaudeBot, and PerplexityBot are all welcome. there's an llms.txt at the root. the schema is clean. by the standard GEO checklist, vaulto is a model student.

and the visibility score is still zero.

that's the counter-narrative most GEO content won't tell you: the technical checklist is necessary but nowhere near sufficient. thousands of words have been written about llms.txt as if it's a magic ticket. it isn't. it controls how easily AI can read your site once it cares. it does nothing to make AI care.

what AI cited instead of us

this is my favorite part of the scan, in a painful way. when Perplexity answered questions about "vaulto", here's what it actually cited:

HashiCorp's documentation for Vault, the secrets manager
a Trustpilot page for a crypto wallet called vaultx
Facebook and Instagram videos about physical vaults and wallets

not one source about us. the model didn't reject vaulto. it has literally never encountered vaulto in a context that matters, so it filled the space with things that sound similar. if your product name collides with bigger entities, you start below zero: AI has to learn that you exist AND that you're distinct from the thing it already knows.

run the same exercise for your own brand and you'll learn more in thirty seconds than from any GEO thinkpiece. the sources AI cites when asked about your category are the exact pages you need to be on.

why technical GEO isn't enough

think of it as two separate problems: the door and the invitation.

on-site work (robots.txt, schema, llms.txt, clear copy) is the door. it determines whether AI models can crawl you, parse you, and quote you accurately. if the door is locked, nothing else matters, so do this work first. it took us about a day.

but AI models decide WHO to recommend based on the invitation: what the rest of the internet says about you. review sites, directories, comparison posts, community threads, news mentions. models are trained on that corpus, and search-grounded models like Perplexity re-read it on every query. when buyers ask "best ai visibility tool", the model synthesizes an answer from sources that discuss tools in that category. if you appear in none of those sources, a perfect door changes nothing.

we built a perfect door into an empty street. nobody had written anything about vaulto anywhere that matters.

the plan we're running to fix it

this is the to-do list our own scanner generated, the same one any low-visibility product gets. we're doing it in public so you can watch whether it works:

get listed where AI looks. launch on Product Hunt, get into the niche directories and "best tools" lists for SEO and AI search. these pages are disproportionately cited by search-grounded models.
publish comparison content. "vaulto vs X" posts, honest ones. AI models lean heavily on comparison content when buyers ask "which tool should i use". if the comparison doesn't exist, the model can't put you in it.
answer real questions in communities. Perplexity weights Reddit and community threads heavily. we'll be useful in places where founders ask about AI visibility, and mention vaulto only when it genuinely answers the question.
build in public. posts like this one. real numbers, including the embarrassing ones, are the only content that's both honest and impossible for competitors to copy.
track the score weekly. vaulto re-scans tracked sites every monday and emails the delta. our own site is on the list. when the 0 moves, i'll publish the before and after, including which actions moved it.

i'm not claiming this plan works yet. that's the point of doing it in public. the score is 0 today, june 10. follow along and judge the results.

check your own score

if you take one thing from this post: your site being well built tells you nothing about whether AI recommends you. the only way to know is to ask the models.

run a free scan on your site. thirty seconds, no signup. you'll get your score across the AI models we query, the technical audit, who AI recommends instead of you, and the prioritized fix list, the same one we're working through ourselves.

worst case, you find out you're at zero like us. best case, you find out before your competitor does.

I Built an AI Visibility Scanner. It Gave My Own Site 0/100.