Everyone Is an AI Company Now, Which Means Nobody Is

TL;DR: "AI company" has stopped carrying information. 337 S&P 500 companies said "AI" on their spring 2026 earnings calls against a ten-year average of 103, and when I token-counted 19 major SaaS homepages, the median said "AI" 27 times, the loudest said it 106 times, and two direct rivals each called themselves "the only" AI platform for the same market. Grade the verb, not the noun: if a sentence still makes sense with "AI" deleted, the AI isn't doing anything.

I fetched the homepages of 20 well-known SaaS companies on August 6, 2026 and counted the literal token "AI" in each one's visible text. Nineteen loaded (Canva's bot-wall said no). Median: 27 mentions per homepage. Maximum: 106, on Zendesk, which brands itself "the only AI-first service platform" while Intercom, selling into the same support market, brands itself "the only helpdesk designed for the AI Agent era." Both of them are "the only." Box sells "AI-driven e-signature workflows." Asana renamed a dashboard "Your AI Chief of Staff." We have officially hit the point where the term tells you nothing.

This isn't a small-sample grievance. In the spring 2026 window FactSet tracks, 337 S&P 500 companies cited "AI" on their earnings calls, well above the 5-year average of 164 and the 10-year average of 103, and saying it on a call is not the same as booking it: only 10 quarterly filings in eight months attach a revenue number to AI at all. Stanford's 2026 AI Index puts organizational AI adoption at 88%. When nearly everyone can truthfully say it, saying it is not a differentiator. It's throat-clearing.

What is the tell?

The tell is the verb. Real AI companies describe what the model does. Fake ones describe what the category feels like. Watch the verbs.

• "Our AI drafts the first-pass reply and flags the 3 accounts most likely to churn." That's a claim. I can test it.
• "AI-powered intelligence that transforms how teams collaborate." That's a smell. There's no object, no output, nothing to measure.

If a sentence about AI still makes sense with "AI" deleted, the AI isn't doing anything. Try it on your own landing page. It's a humbling exercise.

And the count has a beautiful counter-signal hiding in it: the only zero in my sample is Calendly, which ships an AI meeting notetaker and just calls it "Notetaker." Figma, at five mentions, calls its product an "intelligent canvas." The companies with the most confident products are increasingly the quietest about the ingredient, the way nobody markets "database-powered" software.

Why did this happen?

Two forces collided. First, boards started asking every CEO "what's your AI strategy," and "we bolted GPT onto search" became a survival answer. Second, the cost of sounding like an AI company dropped to zero while the cost of being one (data pipelines, evals, retrieval that doesn't hallucinate your pricing) stayed high. When the gap between claim and capability is that wide, marketing fills it.

The hype cycle doesn't punish exaggeration. It punishes silence. So everyone talks.

That's how you get a homepage saying "AI that actually showed up to work," a footer link reading "Hey AI, learn about us" addressed to the crawlers, and a nav where the product line is just "AI." All real examples from the count.

How do you vet an "AI company"?

With a receipts test. Here's the one I now run before I take a meeting:

1. Ask for one number. "What's your model doing that a good if-statement couldn't?" A real team answers in a sentence. A fake one pivots to vision.
2. Ask what breaks. Every genuine AI product has a failure mode the team can describe precisely. If they claim it never hallucinates, they haven't shipped to real users, and their status page is the cheapest place to check the claim, though most AI vendors' uptime numbers do not survive their own incident logs.
3. Ask about evals. "How do you know a new model version is better?" If the answer is "vibes" or "the demo looked good," that's your answer.

I hold the tools in my own stack to the same bar. For Reachium, the platform I build on and disclose, the question was never "is it AI." Everything is. The question was whether the lift is measured and published, and the receipt that survives the test is its connection-notes analysis: a 180,155-request denominator, a stated cohort rule, and a specific finding, 47.80% of accepted requests replying with a note versus 25.54% without. That's the standard: a denominator you can check, not an adjective you can't.

What this means for operators

If you're building, stop competing on the word. You can't win it; the median homepage already says it 27 times. Compete on the thing the word is supposed to point at:

• Ship a benchmark your competitors can't match and put the methodology in public.
• Name your failure modes before a customer finds them.
• Replace one adjective on your homepage with one number this week.

If you're buying, treat "AI-powered" as noise floor. It's the new "cloud-based": technically true, informationally empty, and increasingly not something you can decline, because of 14 SaaS suites with a clear answer, 12 sell no AI-free tier at any price. The signal is downstream: what does it output, how often is it wrong, and what happens when it is.

Is a correction coming?

It is, because labels inflate until they collapse. "Big data" meant something in 2012 and nothing by 2016. "Web3" had a shorter half-life. "AI company" is on the same curve, and I'd guess we're 12-18 months from the term becoming a mild liability in enterprise sales: the moment a CIO hears it and reaches for the receipts instead of the checkbook.

The companies that survive that correction won't be the ones who said "AI" the most; my count already knows who those are. They'll be the ones who, when the word stopped meaning anything, still had a product that did the thing. Everyone is an AI company now. Soon, that'll be the least interesting fact about any of them.

No hype. Just receipts. Start collecting yours.