The BRIEF Method: get Claude Fable to finish the job

Claude working from a half-explained situation guesses, and guesses badly. Point it at the real thing first and make it read before it plans anything.

Here is [your website / CRM / content folder / process doc]. Read all of
it before you say anything back. Then tell me in three sentences what it
currently does, so I know you understood it correctly.

Here is my clinic website, four-chair-dental.com. Read all of it before
you say anything back. Then tell me in three sentences what it currently
does, so I know you understood it correctly.

Result: Claude's first response becomes a summary of reality, not a guess, so you catch a misunderstanding in 30 seconds instead of after an hour of wrong work.

"Make it better" is not checkable. Neither is a task if you hand over your own step list. Claude will follow your steps even when a better path exists two moves away.

New goal: [a result you could check by looking at it]. Not "improve X",
say what a visitor, customer, or you would see or do differently when
this is done. Turn this into 3-5 specific, checkable criteria and read
them back to me before you touch anything.

New goal: a first-time visitor can book an appointment online in under
a minute, without calling the front desk. Turn this into 3-5 specific,
checkable criteria and read them back to me before you touch anything.

Result: you get a checklist you can verify point by point, instead of a vague "does this look better?" conversation at the end.

If you hand Claude your own step-by-step plan, it executes your plan, including your plan's mistakes. Leave the "how" open and it will often find a shorter route than the one you had in mind.

Figure out the best way to hit the goal above yourself. I am not going
to give you the steps. Make the calls on anything I have not explicitly
said is mine to decide.

Figure out the best way to hit that booking goal yourself. I am not
going to tell you which calendar widget to use or where to put the
button. Make those calls yourself. The one thing that is mine to decide:
which insurance providers we list, do not add or remove any without
asking me.

Result: on implementations I run, this single line is usually what separates "did exactly what I said" from "solved the real problem I had."

An AI that says "done" and an AI that shows you a working before/after are not the same level of trust. Ask for proof at every step, not only at the end.

After each change, prove it works: open it, run it, or test it with real
data, and show me the before and after side by side. Do not tell me it
works. Show me.

After you change the booking flow, prove it works: click through it
yourself as if you were a patient, take a screenshot before and after,
and time how long the booking takes. Do not tell me it is faster, show
me the timer.

Result: you stop discovering broken changes a week later. You see proof the same day it happens.

You do not want to approve every small decision, that is slower than doing it yourself. You do want a stop-sign before anything expensive, public, or hard to undo.

Check in with me before any decision that is expensive, customer-facing,
or hard to reverse. Everything else, use your judgment and keep moving.

Check in with me before you publish anything live on the site, or before
you touch the payment or insurance settings. Everything else on the
booking page layout, use your judgment and keep moving.

Result: you get the speed of full autonomy on the 90% of decisions that do not matter, and a checkpoint on the 10% that do.