Most useful Curated answers don't come from the first prompt. They come from the second or third — a follow-up that narrows the question, fixes a wrong assumption, or changes the format. This article walks through the follow-up patterns that work best.
Time to read: ~5 minutes
Why follow-ups beat re-prompts
Every conversation in a Space carries context: the layers in the current Map, the prior question, the answer Curated just gave. When you write a follow-up, Curated builds on that context instead of starting fresh.
Two prompts:
"Find competitor coffee shops within a 10-minute walk of our flagship store."
"Now only ones open before 7am."
The second prompt would be meaningless on its own — what shops? what area? But as a follow-up, it's exact. Curated knows the universe is competitor coffee shops in the walk-buffer; it just adds an attribute filter.
Common follow-up patterns
1. Narrow or expand scope
| Goal | Example | |---|---| | Narrow geography | "Within 5 miles instead of 10."* | | Add a filter | "Only those opened in the last 24 months."* | | Remove a filter | "Ignore foot traffic, just show all of them."* | | Switch to a different region | "Same question but for Travis County, TX."* |
2. Change the output format
| Goal | Example | |---|---| | Get a table | "Show this as a table with name, address, and distance."* | | Get a one-pager | "Summarize this as a one-page PDF brief."* | | Get marketing copy | "Write 3 Instagram captions targeting parents in this area."* | | Get a chart | "Compare these by foot traffic — bar chart."* |
3. Switch sources
| Goal | Example | |---|---| | Use a specific layer | "Run the same analysis but on the Demographics 2024 layer instead of 2020."* | | Use the web | "Search the web for recent zoning changes in this area."* | | Use your own data | "Now use the file I just uploaded as the source."* |
4. Fix a wrong answer
This is where follow-ups shine. Instead of giving up or rewriting from scratch:
| Problem | Follow-up | |---|---| | Wrong layer chosen | "That used the 2020 Census. Re-run with the 2024 Tapestry layer."* | | Wrong jurisdiction | "This is for Texas, not California. Try Texas."* | | Hallucinated number | "That number doesn't match what I see. Show me the source."* | | Missing context | "You forgot that drive-time is more important than walk-time for this use case."* |
A worked example
Initial prompt:
"Find retail centers in Austin with anchor tenants that have signed new leases in the last 24 months."
Curated answers with a map highlight and a list of 18 centers. Good start, too broad.
Follow-up 1 — narrow:
"Only centers larger than 50,000 square feet."
Curated narrows to 6.
Follow-up 2 — switch format:
"Show these as a comparison table with anchor name, square footage, lease date, and tapestry segment of the surrounding ZIP."
Curated produces a 6-row table.
Follow-up 3 — pivot:
"For the top 3 by tapestry score, generate a one-paragraph PlaceStory each."
Curated produces 3 narrative briefs as artifacts.
Total elapsed: ~6 minutes. Same workflow without follow-ups would have been four separate prompts, four context-resets, and probably four trips through wrong assumptions.
When to start fresh instead
Use a new conversation (in a new or same Space) when:
- You're switching to an unrelated topic — Curated will eventually contradict itself if you keep piling on
- You want a clean baseline answer free of any drift from prior turns
- You're showing the work to someone — a fresh thread is easier to read than a 20-message conversation
A good rule: if your follow-ups have been about the same subject for more than ~10 turns, the thread is probably fine. If you've drifted to a new subject, start a new conversation.
Related
- Sign In and Ask Your First Question
- Using Voice Mode
- Understanding Spaces, Artifacts, and Maps