Portfolio AI-visibility auditWhen a buyer asks AI for the best tools in their category, 37% of this portfolio simply doesn't exist.
Seven of nineteen companies were named by no engine, for no prompt. Their competitors were named every time. Here is the receipt.
Nineteen companies, seven chances each to appear.
Every company got the same test: three real buyer prompts, run across three engines, scored on how many of seven cells named them. Ten of nineteen scored one or zero. Not one deep-tech or invented-category company cleared the middle.
One company, every check, and the evidence.
Every cell we ran and every check we made on a single company, with the proof behind each one. Open any row to see exactly what we saw.
The category Huskeys named itself after, asked on Perplexity, logged out
"The best edge security management platforms right now are Zscaler Zero Trust Cloud, Prisma Access, Cisco Umbrella, Netskope One Platform, Check Point SASE, Todyl Security Platform, Cato SASE Cloud, and FortiSASE. G2's July 2026 cloud edge security category highlights those products."
The companynot mentionedTwo things have to be true before an engine says your name.
One of them is on your website. The other one is not. Watch four companies from this portfolio land on the grid, one square at a time, and the shape of the answer becomes obvious.
Zorro4
Obligo5Huskeys0
Dtect0Two things decide whether an engine says your name.
Whether your own page leads with the words a buyer types. And whether a page you do not own says your name back. Only one of those is on your website. Here are four companies from this portfolio, one for each square.
Huskeys invented the words, and nobody repeats them.
Its headline reads "Got WAF? We Make it Work." Its category, "Edge Security Management," is one it coined, and that exact phrase already belongs to a G2 category full of SASE vendors. No ranking carries it. No press covers it. Zero out of seven, and the obvious conclusion is that it should go and fix its words.
Dtect did fix the words. It scored exactly the same.
Its headline is the plainest in the portfolio: prevent financial fraud with deepfake detection. That is close to word for word what a bank would type. On page quality it is the best company in this audit. It has no ranking, no directory entry and no Wikipedia page, and it scored zero out of seven. The same as Huskeys.
Zorro has a vaguer page than Dtect, and four times the score.
Its headline is "Meet the next generation of employee benefits," which never says ICHRA, the term its buyers actually search. The word appears in its title tag, but not where a reader or a model meets the page first. It carries no structured data at all. But Zorro sits inside the "best ICHRA platforms" articles, and the engines quote those articles back. Four out of seven.
Obligo does both, and tops the whole portfolio.
A headline that states the category outright, "The New Standard For Security Deposits," and a place inside the US deposit-alternative rankings the engines read. Five out of seven, the highest score of the nineteen, and the only company here above four.
Along the bottom, nothing changed. Up the side, everything did.
Fixing the words moved Huskeys to Dtect, and the score stayed at zero. Being cited moved Huskeys to Zorro, and the score went to four. In this cohort no company scored above 2 without a third-party page carrying it, and the five highest all sit in a ranking or carry real press. The words are not useless, they decide which question you are eligible to win. The citations decide whether you win it. That is the uncomfortable half, because it is the half that is not a website change. One company, bananaz, reached 3 on unusually specific wording alone. It is the only one that did.
How a company actually gets named.
If being cited is the lever, the practical question is what counts as a citation. Two things reliably earned one here, and one thing can quietly disqualify a company before either matters.
A profile in a directory is not the same as a place in a ranking.
This is the distinction that decides scores. Safebooks and Juno Journey both have a G2 profile, and both scored 0. Existing in a database does nothing, because an engine answering "best X" quotes the article that ranks vendors, not the directory that lists them. Every company here that sits inside a real "best X" ranking scored 3 or higher.
A Calcalist byline shows up inside AI answers.
When Google's AI Overview surfaced an early-stage company from this portfolio, it very often cited the same two sources: Calcalist and CTech. OneStep, Quantum Source and Remondo all appear in AI answers specifically because the Israeli tech press wrote them up and the engine trusts the write-up.
One site is a blank page to any crawler that skips JavaScript.
Robots.txt is wide open across all nineteen, so nothing is deliberately blocked. But one site is client-side rendered, and its raw HTML is nearly empty, so a crawler that does not run JavaScript sees a blank shell where a browser sees a full page. That is not a ranking problem or a naming problem. It is a company that cannot be read at all.
What the audit turned up along the way.
Small things you only find by opening every site and running every query by hand. A few are fixes worth a founder's afternoon.
One entry is out of date.
The portfolio page still lists Mentee Robotics as active although it exited to Mobileye for $900M. We excluded it from the cohort for that reason, so the 37% is calculated on companies that are genuinely still early-stage.
One company has already rebuilt its page.
Safebooks now has a real headline and product structured data, neither of which it had when we scored it. That makes it the cleanest natural experiment in the portfolio: if the words alone were the lever, its score should climb. On the evidence here, it will not climb far until a page it does not own names it.
One site is walled off by its own CDN.
MedOrion sits behind an aggressive Cloudflare challenge that turns away any non-JavaScript client, so we had to read the page in a real browser. A risk worth watching, though Google still indexed it, so it is not the cause of its score.
One Wikipedia page. One ChatGPT mention. Same company.
OneStep is the only company of nineteen with its own Wikipedia article, and the only early-stage company ChatGPT named without being led there. The citation lever, in a single data point.
One company got there on words alone.
bananaz scored 3 with very little off-site presence, purely by owning an unusually specific phrase, "AI for mechanical engineers." It is the clearest exception to the finding above, and worth saying plainly: a narrow enough niche can carry a company on its own.
Six companies publish an llms.txt.
A file that maps a site's content for language models, and six of nineteen already have one, including Huskeys. No engine officially consumes it yet, so it earned nobody a single point. Worth knowing before anyone sells it as a fix.
Nineteen diagnoses. Two or three fixes each.
Every company, what appeared instead of it, and the specific, mostly-free changes that would move its score. Generous on the diagnosis, honest about what sits outside a website's control.
How this was run.
Every prompt was written and locked before a single query ran, so nothing could be cherry-picked after seeing results. Each company was scored the same mechanical way. The point was to build something a founder can check, and repeat.
Seven of these companies are invisible to the engine their next buyer is already asking. None of them know it yet.
The diagnosis is free. The fix is a conversation. → idovadavker.com