AI integrations
Almost every “how to connect X and Y” article documents exactly one route, usually the vendor's own, and presents it as the answer. That is why so many of these setups end in an afternoon lost to a connection that worked perfectly and did the wrong thing. Between any two products there are normally three or four genuinely different routes, and they differ on the one axis nobody publishes: which direction the data moves. These guides compare all of them first, then commit to one recommendation.
ChatGPT + Slack
AI assistants × team chatTwo opposite things get called this. Talking to ChatGPT inside Slack is a five-minute app install. Letting ChatGPT search your Slack history is a business-tier connector plus an admin approval — and no configuration bridges the gap.
ChatGPT + Notion
AI assistants × docs and wikisThe connector reads Notion and can never write to it. Anyone waiting for ChatGPT to file a page through that route is waiting for something it cannot do; writes need MCP, middleware or the API.
Claude + Google Drive
AI assistants × cloud storageRead-only, no sync, and blockable by a Google Workspace admin regardless of your Claude plan — which is why authorisation failing on a work account is a domain policy rather than a bug.
The four routes, in general
Once you can name these, most integration questions answer themselves before you open a settings page.
- The first-party connector. Fastest, free, nothing to maintain — and very often read-only. Excellent when you want to ask about data that lives elsewhere, useless when you want something written back.
- An MCP server. The route that gets you two-way access without writing an integration, because the model calls real tools. The catch is client support, which varies by app and plan, and that write scope plus a loose instruction is genuinely capable of damage.
- No-code middleware. Zapier, Make and friends. The only route that fires with nobody present, which makes it the answer for every “every morning, automatically” request. Watch the task meter and never let an automation write into the folder or channel it watches.
- A script against both APIs. Cheapest per operation by a wide margin, the only sane option for bulk work, and the most expensive in hours. Correct for two thousand rows, absurd for one summary a day.
Three things true of nearly every connector
- Read-only is the default, and it is rarely stated. If a page does not explicitly say the AI can write, assume it cannot. Confirming this before you start saves more wasted time than anything else on this page.
- Sharing scope decides answer quality. These connectors see only what you shared, and they do not know what they are missing — so a gap in sharing arrives as a confident “that isn't documented” rather than an error.
- An admin on either side can end it. Managed Google Workspace domains and restricted Slack workspaces block third-party apps routinely, and the failure surfaces as a broken authorisation rather than a policy message. Check the policy before debugging the product.
How these differ from our other guides
- Comparisons and alternatives help while you're still choosing a tool.
- Pricing pages tell you what a plan costs, and limits pages what you get before it stops.
- Switch guides cover leaving one tool for another. Stacks cover assembling a workflow from several.
- Integration guides are for after all of that — you own both products and now want them talking to each other.
People ask an assistant how to connect two tools before they open any docs.
Integration questions are asked in a chat window now, and the answer names whichever tools the assistant can actually read about. If your product is the missing connector in one of these chains, find out whether ChatGPT, Claude and Perplexity mention it.
📬 Get the best new AI tools delivered weekly
One concise email with fresh launches, trending picks, and featured standouts.