Use Windsor.ai with ChatGPT to investigate your marketing results

Choose the first question before connecting everything

Windsor.ai connects marketing and business data to reporting destinations and AI tools. A manageable starting question is: which campaigns explain the change in enquiries between two completed weeks? Begin with one advertising account whose reporting you understand. Add analytics or CRM data after you can reconcile the first source.

Write down what an enquiry means for this exercise. A form submission, a qualified lead, and a booked consultation are different outcomes. Ask the assistant to preserve those labels throughout the analysis. Otherwise, a confident recommendation can hide a mismatch in what you are counting.

Connect the source and identify the account

For Google Ads, Windsor's setup guide directs you to find Google Ads under Data Sources, sign in with the appropriate Google account, review access, and select the advertising accounts to use. Reporting requires access to those accounts; an organization may need to grant it first. Use the official connector guide for other sources.

Record the account ID as well as its friendly name. Similar names are common when managing several brands. Keep the first query restricted to that account. Windsor's API documentation notes that an unfiltered source query can include every connected account, so an explicit selection prevents an avoidable reporting mix-up.

Connect ChatGPT through the published setup flow

Windsor's ChatGPT guide describes connecting at least one source, opening its ChatGPT app page, choosing Connect, completing authorization, and starting a chat with the connector enabled. It does not require you to preselect fields in Windsor for this integration. Check availability and your organization's app policy if the option is missing.

Windsor also publishes a remote MCP endpoint at https://mcp.windsor.ai/ with OAuth support for compatible clients. Follow the setup guide for your client. Use a fresh connection flow through the official service; do not put account tokens, personal authorization links, or API keys into a reusable prompt or public tutorial.

Select fields and dates deliberately

Ask the assistant to discover the connected accounts and valid field definitions before fetching data. Start with campaign ID, campaign name, date, spend, clicks, and your chosen conversion metric. Add value only if you know what it represents. Windsor's tools support account filters, date ranges, field discovery, and source-specific options.

Use explicit start and end dates for two complete periods of equal length. Request the returned account IDs, row count, retrieval time, and any missing fields. Keep raw totals visible. Calculate cost per enquiry from total spend divided by total enquiries; averaging campaign-level costs can produce a misleading overall figure. With zero enquiries, report the cost per enquiry as undefined.

Reconcile before recommending changes

Check the same period in the source platform. Confirm account timezone, currency, conversion definition, attribution settings, and whether dates represent ad interactions or conversions. Google's documentation explains that reporting views can differ in how conversion credit and time are assigned. Do not add different currencies or platform-attributed conversions together without a defensible normalization or deduplication method.

A fictional example: if a campaign spends 240 units of currency and records eight enquiries, its cost per enquiry is 30. That arithmetic does not establish lead quality, incremental impact, or profitability. Ask what evidence is missing before increasing its budget. Treat sudden changes as investigation leads, especially when the recent period may still be incomplete.

Keep the first workflow focused on analysis

Windsor supports write actions on some connected platforms, including campaign and budget changes. For this exercise, request read operations only and disable write tools where your client allows it. A proposed change should show the evidence and exact affected account before you consider executing it.

Review current Windsor source, account, history, and refresh limits, plus any costs or access requirements of the destination and source services. Do not assume a trial includes your ongoing configuration. Once the analysis agrees with source totals, save the brief and repeat it. Schedule it only after you know how partial data, expired access, and query failures will be reported.

Prompts to put it into practice

Replace the brackets with your context. Use information you have permission to share.

Using Windsor read operations only, inspect account [ID] for [start date–end date] and [comparison start–end]. Discover valid fields first. Return campaign-level spend, clicks, and [exact enquiry metric], plus raw totals, account currency, timezone if available, and retrieval time. Flag missing fields or incomplete results. Rank campaigns by contribution to the change in enquiry count, including low-spend campaigns. Calculate overall cost per enquiry from totals. Separate facts, plausible explanations, and questions needing evidence. Do not change campaigns, budgets, audiences, settings, or exports.

Compare the definitions behind these advertising and analytics results. Identify differences in account scope, date boundaries, timezone, currency, attribution, event names, and freshness. Tell me which comparisons are valid, which need adjustment, and what cannot be concluded. Do not force the totals to match.