Claude can already grade how you use AI, and a more detailed AI Fluency Scorecard was spotted in testing. Here’s what it looks like, all 11 habits it checks, and how to score yourself today.
In May 2026, TestingCatalog spotted an AI Fluency Scorecard being tested in Claude’s settings. Per the report, it looks at your activity across Chat, Cowork and Claude Code, checks 11 observable behaviors, and gives you a score like 7.5 out of 11 plus where to improve. Anthropic hasn’t announced that scorecard or a launch date.
What Anthropic did ship is Reflect (beta, July 2026). Go to Settings, then Reflect, on the web or desktop app. It recaps your last 1 to 12 months of chats and breaks your habits down across Delegation, Description, Discernment and Diligence, with examples and tips. It works on Free, Pro and Max, and Memory has to be on. It doesn’t give a score out of 11.
The 11 behaviors aren’t new. They come from Anthropic’s AI Fluency Index (February 2026), which studied 9,830 Claude conversations. So you can learn them now.
My illustration, built from the leaked scoring prompt TestingCatalog published. The real design may look different when it ships.
Delegation is what you get right before the AI starts working.
01
Clarifies goals
You say what you’re trying to accomplish before asking for help. Strong example. "I need a 150-word cold email that gets a busy founder to book a 15-minute call." Weak is "write me an email."
To level up, put the outcome and what done looks like in your first sentence.
02
Consults on approach
You ask how to tackle something before asking for the finished work. Strong example. "What are 3 ways to structure this launch plan? Which would you pick?"
To level up, get options first on anything that matters, and choose before you say go.
Description is how clearly you frame the request. Vague in, vague out.
03
Defines audience
You say who the output is for. Strong example. "For busy founders who skim on their phone."
To level up, add what the reader already knows and what they care about.
04
Specifies format
You name the shape you want, like a table, an email or a word count.
To level up, say the format before it writes, not after.
05
Communicates tone
You describe the voice you want.
To level up, name a tone and one thing to avoid, like "confident, no corporate jargon."
06
Builds iteratively
You refine with follow-ups instead of taking the first answer.
To level up, say "this is 70% there, here’s what’s off" instead of starting over.
07
Provides examples
You share an example or reference of what good looks like.
To level up, paste one real example you want it to match. It’s one of the biggest quality boosts there is.
08
Sets interaction
You tell the AI how to behave, what role to take or how to work with you. Strong example. "Act as a skeptical CFO and ask me one question at a time."
To level up, start important chats with a role and a working style.
Discernment is quality control, and it’s where most people lose points. In Anthropic’s data, people iterated in 85.7% of conversations, but they questioned Claude’s reasoning far less often. When Claude produced a finished artifact like a document or app, people questioned it even less.
09
Checks facts
You question or verify claims, numbers and citations.
To level up, ask "what here should I double-check, and what are you least sure about?"
10
Notices reasoning
You push back when the logic seems off or it dodged the hard part.
To level up, ask it to argue the opposite case, then see which side holds.
11
Recognizes context
You offer context the AI couldn’t know on its own. Strong example. "Heads up, our last launch flopped because of pricing, so factor that in."
To level up, before you hit send, ask yourself what someone new to this would be missing.
Fix One Thing This Week
Start questioning the output. Discernment is the rarest group of habits, so it has the most upside on the scorecard and at work.
Want the 11-habit score now? Open a recent Claude chat where you did real work and paste this at the bottom.
Once you know the habits, you don’t have to think about them every time. This skeleton bakes all three groups into one prompt you can reuse for anything that matters.