Write better questions
How to ask your AI questions that get precise answers from your accounts - exact dates, naming the account, asking for tables, and checking the result.
Your AI reads your accounts live through Plugsight, but it still has to work out what you mean. The more precise the question, the fewer guesses it makes. These habits help most.
Say which data to use
AI apps sometimes answer from general knowledge instead of your accounts. Make it clear:
- "Using Plugsight, show last week's Google Ads spend."
- "From my Zendesk, how many tickets were created yesterday?"
Also check the Plugsight connector is switched on in that chat (in Claude: + → Connectors; in ChatGPT: + next to the message box).
Give a date range
Without one, your AI picks its own, and marketing tools default to the last 30 days.
| Vague | Precise |
|---|---|
| "recently" | "in the last 7 days" |
| "this month vs last" | "this month to date, compared with the same days last month" |
| "last quarter" | "from 2026-07-01 to 2026-09-30" |
| "this week" | "Monday 9 November to Sunday 15 November 2026" |
Exact dates (YYYY-MM-DD) are the most reliable, especially when you'll repeat the report. For how presets like "last 30 days" are worked out, see Get accurate numbers from marketing data.
Name the account
When the server reaches more than one account of a kind, say which one:
- "In my work Gmail, find the latest invoice from Northwind."
- "For Bright Dental's Google Ads account, which search terms wasted money?"
- "On the shop's GA4 property, sessions by channel last month."
Each connection also has a short name, shown under its label on Connections, which you can use too, for example "check [short name] for new messages". Not sure what's there? Ask "List my Plugsight accounts" or "List my marketing accounts".
Or ask across everything on purpose: "in any of my inboxes", "across all client accounts".
Ask for the shape you want
Tell your AI how to lay out the answer:
- "Show it as a table with columns: campaign, cost, conversions, CPA, change vs last month."
- "Top 10 only, sorted by cost, highest first."
- "One row per account." (keeps accounts and currencies apart)
- "Under 150 words, as bullet points."
- "Include the ticket numbers" or "include the order numbers", so you can check them.
One job per question
Big questions are easier to check in steps:
- "List my Google Ads accounts."
- "For the Acme account, cost and conversions by campaign last month."
- "Now the same for the month before, and show the change."
- "Which of these campaigns had the biggest drop, and why might that be?"
Some data can't be combined in one request anyway, for example Google Ads search terms with quality score. Your AI is told when that happens and splits the question.
Use the ready-made analyses
For marketing, Plugsight has deep dives that run several queries at once. Ask for them by name:
- "Run a wasted spend report."
- "Run an SEO opportunities report for this site."
- "Give me an account overview, this month against last month."
- "Run a period comparison of cost and conversions by campaign against the same period last year."
See How your AI analyses marketing data for the full list.
Be exact about changes
For anything that sends or changes, say exactly what you want, and read the preview before you say yes:
- "Draft a reply to Sam, from my work inbox, saying Thursday 2pm works. Don't send it yet."
- "Set ticket 4821 to pending and tag it 'refund'."
Your AI shows the final email, ticket change or booking and waits for your approval. See Approvals in the chat.
Check the answer
- Ask "Which accounts and dates did you use?" Every result says where it came from.
- Ask "Were there any data notes?" Plugsight adds notes when providers hold data back.
- Spot-check one number in the provider's own interface.
Save questions that work
Keep the questions you reuse in a note: your morning inbox triage, your monthly report, your support review. Same questions, same order, comparable answers.