PromptCorrectly
PacksBundlesLibraryGamesBlog
Sign inStart free
PromptCorrectlyBLOG
← All articles
STRATEGY · 10 min read

AI in 2026: 900 Million People Use It Every Week — So Why Do Most Results Disappoint?

promptcorrectly.com · Updated 2026-11-16

Almost everyone uses AI now, but most of what it produces at work still disappoints. ChatGPT alone reached 900 million weekly users in early 2026, and 88% of companies use AI somewhere. Yet one widely cited MIT study found that 95% of organisations saw no measurable profit from their generative AI spending, and four in ten US workers say they received sloppy AI-made work in the past month. The research points to the same cause again and again: the AI is not given enough context. That is a prompting problem — and it is fixable.

This article pulls together the most-cited numbers from 2025 and 2026, explains what they actually measure, and turns them into a short list of changes you can make today.

The adoption numbers: AI is now normal

Individuals. OpenAI said in February 2026 that ChatGPT had reached 900 million weekly active users and 50 million paying subscribers, up from 800 million weekly users in October 2025 (TechCrunch, 27 Feb 2026). That is one product. Add Gemini, Claude, Copilot and the AI built into phones and office software, and AI use is close to universal among people who work at a screen.

Companies. McKinsey's global survey, published in November 2025 with 1,993 respondents in 105 countries, found that 88% of organisations now use AI regularly in at least one business function, up from 78% a year earlier (McKinsey, The state of AI in 2025).

Workers bringing their own. In the MIT NANDA study of enterprise AI, only about 40% of companies said they had bought an official AI subscription — but people at more than 90% of the companies reported using personal AI tools for work anyway (coverage in Tom's Hardware). The tools are already on everyone's desk, whether IT handed them out or not.

So access is no longer the bottleneck. Something else is.

The results numbers: most of it is not paying off yet

Few companies see real returns. The same McKinsey survey found most organisations still in pilot or experimentation mode, and only about 39% could point to any measurable effect of AI on their bottom line (McKinsey).

The "95%" study. MIT NANDA's report The GenAI Divide: State of AI in Business 2025 estimated that, despite $30–40 billion of enterprise spending on generative AI, 95% of organisations were getting no measurable return on their profit and loss (Virtualization Review). The number has been quoted everywhere, so it is worth being precise about it: it comes from one study, built from interviews, a survey and a review of public deployments, and it measures profit impact, not whether people found the tools useful. Even so, the direction matches McKinsey.

"Workslop." Researchers at BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 full-time US workers for Harvard Business Review in September 2025. About 40% said they had received "workslop" in the past month — AI-made work that looks finished but is too vague, generic or wrong to use. Each case took close to two hours to sort out, which the researchers estimated at about $186 per worker per month in lost time (Allwork.Space summary). Those cost figures are self-reported estimates, not measured losses — but two hours is two hours.

Training lags use. BCG's AI at Work 2025 survey of more than 10,600 workers in 11 countries found that only 36% feel they have been properly trained to use AI, and 18% of regular users had received no training at all (BCG).

What the research says is going wrong

Read these studies side by side and the same pattern appears. The MIT authors call it a "learning gap": most AI set-ups don't keep context, don't remember what they were told, and aren't fitted to how the work is actually done. The workslop researchers describe output that has the shape of good work without the substance. BCG finds people using the tools daily with no one ever showing them how.

None of these reports blame the models. They point at what goes into the model:

  • No context. "Write a follow-up email" — to whom, about what, after what, in what tone? The model fills every gap with the most average possible answer.
  • No standard of good. Nobody said how long, for which reader, what must be included and what must never be. So "done" means "it produced text."
  • No memory between tasks. The same background gets retyped (or forgotten) every session, so quality depends on how patient the person was that day.
  • No review step. The first draft gets forwarded. That is how workslop travels.

In other words, most disappointing AI output is a briefing problem. A capable colleague handed a one-line note would produce something generic too.

Five changes that close most of the gap

You don't need a new tool. These five habits address exactly the failures the studies describe.

1. Give it a role and the situation

Start with who the AI should be and what is going on: "You are an experienced bookkeeper for small cafés. I run one café in Lyon with six staff; it's the end of the quarter." Two sentences remove most of the guesswork. Our guide to role prompting shows why this works.

2. Say what good looks like before it starts

Length, reader, format, must-include facts, things to avoid. "Under 150 words, for a customer who complained, keep the refund policy accurate, no exclamation marks." This is the step most people skip — and it is the difference between a draft and workslop.

3. Use a structure, every time

A good prompt has five parts: Role, Context, Task, Constraints, Output. Writing them in that order makes it almost impossible to forget the important bits. See the RCTCO prompt structure for examples.

4. Write your background once and reuse it

The "learning gap" is partly a memory gap. Keep your business details — name, location, prices, tone, customers — in one place and paste them in, or put them in a Claude Project, a custom GPT or a Gemini Gem so they're there every time.

5. Ask it to check its own work

Before you use an answer, ask: "List anything here you're unsure of or that I should verify." It costs ten seconds and catches the errors that turn into two-hour clean-ups for someone else. More in self-critique prompting.

Where ready-made prompts fit

Writing all five parts from scratch for every task is the reason many people don't. That is the gap a finished prompt fills: someone who knows the job has already written the role, the constraints and the output format, and left blanks for your details.

Every PromptCorrectly pack also includes a start file, 00-GIVE-THIS-TO-YOUR-AI.md. You give it to ChatGPT, Claude or Gemini and it asks for your business details once, remembers them, and fills the blanks in every prompt and skill in the pack — which is habit 4 done for you. If you'd rather start free, there are hundreds of free prompts built the same way.

The short version

  • AI use is close to universal: 900M weekly ChatGPT users, 88% of companies.
  • Results lag: ~39% of companies see any bottom-line effect; one MIT study puts 95% at no measurable profit.
  • About 40% of US workers recently received AI "workslop", costing around two hours each time.
  • Only 36% of workers feel properly trained.
  • The common cause is missing context and missing standards — fix the brief and most of the gap closes.

Questions people ask

How many people use AI in 2026?

OpenAI reported 900 million weekly active ChatGPT users and 50 million paying subscribers in February 2026. McKinsey's 2025 survey found 88% of organisations use AI in at least one business function.

Is it true that 95% of AI projects fail?

That figure comes from MIT NANDA's 2025 report, which found 95% of organisations saw no measurable profit-and-loss return from generative AI. It measures profit impact from one study, not whether people found AI useful, but McKinsey's data points the same way: only about 39% of companies see a bottom-line effect.

What is AI workslop?

Workslop is AI-generated work that looks finished but is too vague, generic or wrong to use, so someone else has to fix it. In a 2025 BetterUp Labs and Stanford survey, about 40% of US workers had received some in the past month, and each case took close to two hours to resolve.

Why are AI results so often disappointing?

Mostly because the AI is given too little context and no standard for a good answer. Adding a role, the situation, clear constraints and an output format — and reusing your background details — fixes most generic output.

Put this into practice

Grab a ready-made prompt from the Library, or a pack for your trade, fill in your details and take it to the AI you already use.

Browse the prompt libraryShop the packs

Keep reading

Why Most People Get Mediocre AI Results (and How to Fix It)

Mediocre AI output is almost never the model's fault. Here are the 6 prompting mistakes that flatten your results, each with a before/after fix.

9 min read

How to Use AI Properly: The Beginner's Guide That Actually Works (2026)

How to use AI properly at work, at school and at home: the 5 habits that separate great results from generic ones, with copy-paste examples for ChatGPT, Claude and Gemini.

11 min read

Same Question, Different Words: How Small Prompt Changes Swing AI Results (15 Swaps That Work)

Research shows formatting alone can swing AI accuracy by up to 76 points. Why a few words matter so much, plus 15 word swaps for better answers.

11 min read
← All articles
promptcorrectly.comFROM VAGUE INTUITION TO STRUCTURED INSIGHTLIBRARY · PACKS · GAMES
PromptCorrectly

The marketplace for AI prompts, skills and packs. Ready-to-use prompts, installable AI skills and trade packs for ChatGPT, Claude and Gemini. Browse free, or buy a ready pack from €4 — pay once, keep it.

Product
  • Packs
  • Bundles
  • Skills
  • Prompts
  • Free prompts
  • Free skills
  • Games
  • A–Z index
  • Every job we cover
  • Made for your AI
  • Basket
  • Search
Resources
  • Prompts by field
  • All prompts A–Z
  • How to prompt correctly
  • Community uploads
  • AI glossary
Company
  • About
  • Contact
  • FAQ
  • Blog
© 2026 PromptCorrectly · The marketplace for AI prompts, skills and packs.
TermsPrivacyRefundDisclaimerAcceptable useCookies