AI can process information — but it does not automatically connect to it
Power Automate can use AI to interpret, classify, extract or summarise information. But the AI step still needs the information to be supplied to it. Reaching an external website or business system is a separate integration problem.
Last reviewed 2026-09-05
Processing vs connectivity
| Need | What solves it |
|---|---|
| Extract the total and supplier from an invoice already received | AI Builder invoice/document processing |
| Classify text already present in the flow | AI Builder classification or prompt |
| Summarise text already supplied to the flow | AI Builder / prompt action |
| Read data from a Microsoft 365 service | Use the relevant Microsoft 365 connector |
| Call an external API or fetch a generic web resource | Use an HTTP/API action, a suitable connector or a custom connector |
| Interact with a web application that has no suitable API | Consider browser/RPA automation, with stronger governance and maintenance |
Example: monitoring the DfE website automatically
| Step | Typical Power Automate design |
|---|---|
| 1. Schedule | Run the flow daily or weekly. |
| 2. Fetch | Retrieve the target public page or API using an appropriate connector/HTTP request. |
| 3. Compare | Compare the current content, timestamp or stored representation with the previous version. |
| 4. Interpret — optional | If needed, pass the changed content to an AI prompt to explain or categorise the change. |
| 5. Notify | Send a Teams/email alert with the source link and the detected difference. |
| 6. Record | Store the date, source, result and review status so there is an audit trail. |
Once you want the check to happen without a person remembering to ask, you have moved from AI assistance into automation and integration.
External applications need an integration route
For an external system, first ask whether Microsoft already provides a connector. If not, check whether the system exposes an API that can be called with HTTP or a custom connector. If neither exists, browser/RPA automation may be possible, but it is usually more brittle and needs stronger ownership, testing and change monitoring.
Related guidance
Sources
- AI Builder in Power Automate overview — checked 2026-09-05
- Use the invoice processing prebuilt model in Power Automate — checked 2026-09-05
- Overview of HTTP connectors in Power Automate — checked 2026-09-05
- HTTP With Microsoft Entra ID connector — checked 2026-09-05
- Add knowledge sources to your declarative agent in Microsoft 365 Copilot — checked 2026-09-05