12 AI Taste Training Examples You Can Adapt Today

Stuck on how to word your feedback? This swipe file gives twelve examples of AI taste training notes, each turned into a rule, across different kinds of small business. They are illustrative samples written for this post, not records from real customers. Change the details to fit your brand. The method comes from Lesson 1.3: Teach AI your taste.

A good rule names one specific thing, says what to do rather than only what to avoid, and can be followed by a stranger.

How to use these examples

Each one has the note you might write, the rule it becomes, and a short reason it works. Copy the pattern, not the words. Then ask your AI to apply the note to everything and save it as a rule.

The Claude Code docs say specific, concise instructions are followed more consistently, and that a few relevant, varied examples improve consistency. These examples follow both ideas.

Examples for image posts

1. Bakery: text outlines Note: "White outlines on text look tacky." Rule: No white outlines on text. Why it works: one visual detail, easy to check.

2. Candle shop: the hero product Note: "The candle is too small." Rule: The candle is always the hero of the image, large and centred. Why it works: it says what to keep, not only what to avoid.

3. Florist: readable text Note: "I cannot read the words on the bouquet photo." Rule: Use a text colour that stands out from the photo. Why it works: it names the goal, which is contrast.

4. Skincare seller: line breaks Note: "One word sits alone on the last line." Rule: Never leave a single word alone on its own line. Why it works: it is literal and checkable.

Examples for layout

5. Coach: where text goes Note: "Text covers her face." Rule: Text goes only on the left or at the bottom of the image. Why it works: it gives positions, so the AI need not guess.

6. Cafe: space around the product Note: "Everything feels crowded." Rule: Leave empty space around the main item. One item per image. Why it works: it replaces a feeling with a count.

Examples for wording

7. Fitness trainer: banned words Rule: Never use the phrase "transform your life". Describe the plan in plain words. Rule: Describe the plan in plain words. Why it works: it lists words and gives an alternative.

8. Accountant: no promises Note: "It sounds like we guarantee savings." Rule: Never promise a specific saving or outcome. Why it works: it covers a whole class of claims. For anything about money or regulation, check the facts against an official source yourself.

9. Photographer: caption openings Note: "Every caption starts with 'Excited to share'." Rule: Open each caption with one plain sentence about the photo. Why it works: it replaces a habit with a pattern.

Examples for voice and process

10. Pet groomer: add a real detail Note: "Sounds like anyone could have written it." Rule: Leave a line in each draft where I add one true story or detail before posting. Why it works: it keeps your voice in the loop. AI posting without your feedback is how work begins to sound generic.

11. Online tutor: apply to everything Note: "You fixed Monday but not Thursday." Rule: When I give feedback, apply it to every piece, not only the ones I mention. Why it works: it stops partial fixes.

12. Any business: check saved rules Note: "The rule file keeps growing." Rule: Each month, list the rules and flag any that clash or no longer apply. Why it works: the docs say large instruction files cause trouble, so pruning helps.

Turning an example into your own

1. Pick the closest exampleChoose one from your niche or the nearest match.
2. Swap the detailsReplace the product, colour or word with your own.
3. Test it on a strangerCould someone new follow it? If not, make it more literal.
4. Save itAsk the AI to store it as a rule and read what it wrote.
5. Check next weekSee if the rule held in new work.

Where each example lives

Layout and wording rules suit a CLAUDE.md or similar memory file. The docs describe CLAUDE.md files you write and auto memory the tool writes, and a /memory command to view them. See Lesson 1.1 for a setup that keeps files between sessions.

Patterns that show up across all twelve

Look at what the good examples share.

• They name a single thing. One rule covers one behaviour.
• They are literal. Positions, colours and words are named, not hinted at.
• They state the keep as well as the stop. Several rules say what to do, not just what to avoid.
• They are short. A line or two each.

Weak versions of the same rules tend to use soft words: nicer, cleaner, more professional. If you catch one in your own list, rewrite it as a position, a colour or a word.

Mistakes to avoid when copying

• Copying a rule that does not fit. A ban on one word may be pointless in your niche.
• Copying too many at once. You will not be able to tell which one worked.
• Skipping the check. After saving, read the stored rule and test it on the next batch.

Building your own list

Start with the top three annoyances from your last week of output. Write each as a note, then a rule, then give one reason. When you have a dozen, you have your own swipe file, built from your own taste, which is the point of the lesson.

Questions people ask

Are these real client rules?

No. They are invented examples written to show wording patterns.

How many should I start with?

Two or three. Add more after a week of checking.

Can I use these with any AI tool?

The wording works anywhere. Saving them as lasting rules needs a tool that keeps instructions between sessions.

Where to go next

See the full process, with notes turning into rules on a week of posts, in Teach AI your taste, Lesson 1.3. For a fill in format for your own list, use a free AI taste training template.