Tips & Best Practices

Structure your grid before processing

Fill in your manual columns first — especially the primary identifier column (company name, person name, topic). This gives the context needed for good results. A row with nothing in it will produce empty or low-quality output.

Recommended setup order:

1. Create the grid and add all columns
2. Add all your rows and fill in the identifier column(s)
3. Configure AI columns with their prompts
4. Process the grid

Chain columns together for multi-step research

Use the output of one AI column as the input for another. This lets you break complex tasks into simple, reliable steps.

Example — Lead Qualification:

1. Company Name — manual
2. Website URL — "Find the official website URL for @Company Name." (URL, web browsing on)
3. Company Description — "Based on @Website URL, write a 2-sentence description of what @Company Name does." (Text, web browsing on)
4. Industry — "Based on @Company Description, what industry is @Company Name in? Choose one: SaaS, Fintech, Healthcare, E-commerce, Other." (Select, web browsing off)
5. Qualification Score — "Given that @Company Name is a @Industry company described as: @Company Description — rate their fit as a customer for a B2B project management tool on a scale of 1 to 10." (Number, web browsing off)

Notice how later columns reference earlier ones, and only the columns that need current web data have web browsing enabled.

Use format types to enforce clean output

If you want a number, set the column to Number and say "return a number only" in the prompt. Without this, the output might be "approximately 500 employees" instead of 500. Enforce both with the format type setting and a clear prompt instruction.

• Want a URL? Use URL format + "return only the URL, no other text"
• Want a yes/no? Use Checkbox format + "answer Yes or No only"
• Want exactly three items? Use Bullets format + "list exactly 3 bullet points"

Enable web browsing only where needed

Web browsing is powerful but slow. Every web-browsing cell makes a live search and reads pages. On a 50-row grid with 4 web-browsing columns, that's 200 web operations.

Strategy: Use web browsing for a small number of foundational columns (website URL, company description, LinkedIn), then build all subsequent analysis on top of those results — without any more web browsing.

Use the grid context to set a persistent lens

If all your columns should be analyzed through the same lens, set it once in the Grid Context (grid settings → Context). This applies to every cell without you repeating it in every prompt.

Example grid context: "This grid analyzes European fintech startups from the perspective of a Series A venture capital fund. All assessments should reflect an institutional investor's perspective."

Then your column prompts stay lean:

• "Summarize what @Company Name does." (No need to re-explain the context)
• "Rate the market opportunity: High, Medium, or Low."

Review low-confidence cells first

After processing, scan for cells with a low confidence score (hover to see it). Cells below ~60–70% confidence are the ones most likely to be wrong or incomplete.

Quick audit workflow:

1. Process the grid
2. Scan for red/error cells first — fix or regenerate
3. Then review low-confidence cells (hover to check the score)
4. Mark reviewed cells with a thumbs up, flag questionable ones with a thumbs down

Use @mentions instead of rewriting context

If you've already generated a company description in Column B, don't re-describe the task in Column C — just reference it:

• Bad Column C prompt: "Research @Company Name and write a sales pitch."
• Better Column C prompt: "Based on this company description: @Company Description — write a 3-sentence sales pitch."

The second version uses already-generated information instead of triggering another research step. It's faster, cheaper, and usually more consistent.

Manually edit without hesitation

Your manual edits are completely safe. Manually-edited cells are never overwritten during batch processing. You can fix a wrong answer, add information that was missed, or completely replace an AI value with your own.

If you later want to regenerate a manually-edited cell, right-click and choose Regenerate — you'll get a confirmation before your edit is overwritten.

Document analysis: upload first, then reference explicitly

Uploaded documents aren't used automatically. You must explicitly @mention a document in a column prompt for it to be included.

Document workflow:

1. Upload documents in grid settings
2. Write column prompts that explicitly mention the documents: "Based on @Contract.pdf, list the key payment terms."
3. Use @All Documents to include everything you've uploaded

Start with a template, then customize

Don't build from scratch if a template gets you 80% there. Start from the closest system template, review the pre-built prompts, delete what you don't need, and refine what you do.

Save your best grids as templates

When you've perfected a workflow — a competitive analysis structure, a lead scoring framework, a document extraction setup — save it as a template before you change anything. This takes 10 seconds and pays off every time you need to run the same workflow again.

Use sharing to deliver results to clients

Instead of exporting to Excel and emailing a file, share a live link. Clients can view your grid in their browser, see all the formatted content, and come back to it anytime. When you update the grid, the link reflects the latest version immediately.