The site ranks. The word “AI” is in the H1, the title tag, the category page and forty blog posts. Search Console shows impressions climbing. And the pipeline from organic is empty, because everything the site ranks for was searched by people comparing tools, writing listicles, or looking for a job. The buyer, the one with a budget and a problem, searched for something else.
The pattern: AI startups rank for the word “AI”, and buyers never type it. The queries that pay describe the job.
“AI” stopped being a keyword
Three years ago “AI” in a query narrowed the field. Today every product in every category has it in the name, so a search engine treats it the way it treats “software” or “platform”: a word that describes almost everything and therefore selects almost nothing.
That has two consequences. The first is that ranking for “AI [category]” puts you next to every competitor with the same word in their name, in a results page that review sites and tool directories already own. The second is that the buyer knows this too. When someone actually needs the job done, they stop typing “AI” and start typing the job.
The three query groups, and where AI startups end up
I split SaaS demand into problem, category and comparison queries in SEO for SaaS. AI startups have a sharper version of the problem: the category group is almost entirely poisoned.
- “AI [category]” queries have the biggest numbers on the keyword report and the lowest buyer intent. Most of them are people building a list of tools, not choosing one. The results pages belong to directories.
- “Best AI tools for [job]” is the same intent with a different word order. Ranking there means being one line in a listicle someone else controls.
- “[Job] automation”, “how to [job] faster”, “[job] without [manual step]” are the queries with money in them. They rarely contain the word AI. They describe an outcome, and the person typing them has the problem right now.
Most AI startup content plans invert this. They chase the first two groups because the volume is visible, and skip the third because it does not look like “their” keyword.
What the engines call you
There is a second version of this trap, and it is quieter. Ask ChatGPT or Perplexity what your company does. If the answer is “an AI platform for businesses”, you have been filed under nothing.
AI answers assemble a description of you from your own site and from what others write about you. When every page of yours leads with “AI-powered”, the engine learns that you are AI-powered and little else. The startups that get named inside answers to “how do I [job]” are the ones whose pages say, in the first sentence, which job they do and for whom. The method for checking this is in how to measure AI search visibility. For an AI startup, the category questions in that method are the ones to watch.
In the keyword research for this site, the “ai visibility tools” cluster alone had 2,100 searches a month and had grown seventy-fold in a year. Every result on the first page was a software directory or a tool vendor. Not one was a company someone would hire.
What to do instead
- Write the product description without the word AI. If the sentence still says what you do and for whom, that is your positioning. If it collapses, the problem is upstream of SEO.
- Build the keyword list from the job, not the technology. Start from what the buyer was doing manually before you existed, and the words they used to complain about it.
- Give the AI terms one page and no more. A single page can hold the “AI [category]” intent for people who search that way. Forty pages on it compete with each other and with the directories.
- Make the first sentence of every page quotable. An engine assembling an answer takes the sentence that answers the question. “AI-powered platform that leverages” is not that sentence.
- Get described correctly by three sources you do not control. A partner page, an analyst note, a customer’s case study. Those decide what the engines call you more than your own homepage does.
This is the content decision that a content strategy is supposed to settle before anything is written. Google’s own guidance on helpful content says the same thing from the other side: pages made primarily to gain rankings are the ones it is trying not to reward.
The test
Take your ten best-ranking queries and ask, for each one, who typed it. If the honest answer is “someone writing a listicle” or “someone comparing tools for a spreadsheet”, the ranking is real and the traffic is not yours.
Then take the ten queries your last five customers would have typed the week before they found you. If none of those appear in your rankings, that is the gap, and it is the whole plan.
If you are an AI startup and the organic numbers look good on paper and empty in the CRM, this is the pattern, and it is what I look for first in AI search optimization work. Tell me what your product does without the word AI, and I will tell you which queries you should own instead.