Enterprise AI adoption: Who’s actually winning?

TL;DR

• The findings come from the Xensam Insights Enterprise AI Adoption Report, H1 2026, based on observed usage across more than one million professional users.

• Microsoft leads enterprise AI adoption on both fronts at once, 93% reach and 91% engagement, a combination no other vendor comes close to.

• OpenAI is still the largest AI assistant by scale, but engagement dropped from 87.7% to 67.9% in three months.

• Anthropic is the only major AI vendor growing reach and engagement together, engagement nearly doubled in six months, from 31.0% to 65.8%.

• Google’s AI reach nearly quadrupled this year, engagement collapsed to 4.7%.

• Across every vendor, desktop AI apps consistently earn far more daily use than their browser-based equivalents.

Four vendors, one very uneven race

You’ve probably seen the same four logos everywhere this year. Microsoft has Copilot. OpenAI has ChatGPT. Anthropic has Claude. Google has Gemini. Every one of them is telling a growth story right now, and on paper, it looks like four vendors racing neck and neck for the same enterprise budget.

Look past the press releases and the picture gets a lot less tidy.

We tracked observed activity from more than one million professional users across corporate devices, not licenses, not surveys, not a single vendor-reported figure. And what that data shows is that reach (how widely a tool spreads) and engagement (how deeply it’s used once it’s there) are moving in wildly different directions depending on who you ask.

Some vendors are winning on both fronts. One is barely winning on either.

That gap is the real story of enterprise AI adoption right now: not who has the most visible product, but who’s built a habit around it. If you’re the one deciding what to renew, expand, or let lapse, keep reading.

The vendor-by-vendor breakdown

Microsoft: the clearest lead in the market, and it’s not close

By June, Microsoft reached 93.0% of organizations while holding 91.2% engagement, easily the strongest combination of breadth and depth anywhere in the dataset.  

That lead is a portfolio play. Microsoft 365 Copilot Desktop alone closed the period at 98.0% engagement, and Microsoft 365 Copilot Desktop Chat grew from 32.2% to 50.3% reach while holding steady engagement in the high 80s.  

Even Microsoft’s weakest layer, Microsoft Copilot [Web], still reached 82.4% by June, even as its engagement slid to 32.0%. When your worst-performing product is still that widely used, you can afford a soft spot.

OpenAI: still enormous, but the momentum has cooled

ChatGPT [Web] remains one of the most-used AI products anywhere in this report, 80.4% reach with 67.7% engagement in June.  

That’s not a small footprint. But watch the trend line, not just the snapshot: engagement peaked at 87.7% in March, dropped to 66.7% by May, and only partially recovered by June.  

ChatGPT Desktop tells a stranger, smaller story of its own, just 6.1% reach, but 100.0% engagement, meaning almost nobody’s adopted it yet, but the ones who have are hooked. OpenAI is maturing, and maturing incumbents get judged differently than fast-growing ones.

Anthropic: the only vendor improving on everything at once

Here’s the number that should catch your attention if you haven’t been tracking Anthropic closely: reach climbed from 1.8% to 16.5% over six months, and engagement climbed right along with it, from 31.0% to 65.8%.  

No other vendor in the report managed to grow both metrics at the same time. Claude Desktop is doing the heavy lifting, holding near-100% engagement for the entire period while reach grew from 2.7% to 11.7%. Claude Code 2 shows the same pattern at a smaller scale. Multiple products are trending the same direction, not one viral hit carrying the whole brand.

Google: the most access, the least loyalty

Gemini’s reach nearly quadrupled this year, 4.8% to 18.0%.  

If you stopped reading there, you’d think Google was having a great year. But engagement moved in the opposite direction, falling from 29.1% to 4.7%, with the sharpest drop landing in May and June. By June, Google holds a slightly higher reach than Anthropic, 18.0% versus 16.5%.  

On engagement, there’s no contest: 4.7% versus 65.8%. More people can open Gemini than ever before. Almost none of them are choosing to.

Making it real

Here’s the pattern that should matter most to you if you’re the one signing off on AI spend: this reach-versus-engagement gap shows up inside every single vendor, between their own desktop and web products. 

Microsoft 365 Copilot Desktop hit 98.0% engagement. Microsoft Copilot [Web] hit 32.0%. Same company. ChatGPT Desktop returned to 100.0% engagement in June. ChatGPT [Web] sat at 67.7%. Same company again.  

Across the entire report, installed and embedded AI tools consistently out-engage their browser-based counterparts, even when the browser version has far more reach. The web is where people try an AI tool once. The desktop is where it earns a permanent place in someone’s workflow. 

If you’re comparing AI tools purely by adoption numbers or license counts, you’re measuring the wrong thing.  

A tool with high reach and low engagement = a line item quietly losing you money.  

A tool with rising reach and rising engagement, the way Anthropic is trending right now, is worth watching closely before your next renewal conversation, not after it. 

What this means for your next AI decision

This is what a million people did on their own devices, month after month, with no survey asking them how they felt about it. 

• Microsoft’s lead is real and structural.  

• OpenAI is still massive but no longer accelerating the way it was.  

• Anthropic is the only vendor improving on every front at once.  

• Google has more access than ever and less to show for it.  

If you’re making a call on enterprise AI adoption, vendor consolidation, or where next year’s budget goes, this is the data you want in front of you before that decision gets made, not after. 

The full Xensam Insights Enterprise AI Adoption Report, H1 2026, breaks all of this down product by product, plus a look at the smaller vendors, Cursor, Perplexity, Svea GPT, starting to show up in the data before they show up in your procurement pipeline. 

Discover more in the full report.