Automation · Migration

Your ChatGPT Zaps broke on 26 August. Only one of the five is a real rebuild

The sunset shipped on schedule, and Zapier retired five ChatGPT actions with it. Most write-ups stopped at the list. Here is our call on each one: three you delete, one you verify, one you actually rebuild.

By Ishan Vats · Founder of IV Consulting, builds AI agents & automations for 150+ teams

Updated 4 Sep 2026 · sunset verified against OpenAI's deprecation list 8 min read Pillar: Workflow automation

My Zap broke, what now? → Price the rebuild first

Zapier ChatGPT (OpenAI) Assistants API Sunset complete

Sunset shipped 26 Aug 2026

OpenAI retired itAssistants API

5 actions affectedYour ChatGPT Zaps

3 actionsDelete

1 actionVerify

1 actionRebuild

No extensionlisted as a past deprecation

Quick answer

On 26 August 2026 OpenAI switched the Assistants API off on schedule, and any Zap still using Zapier's Assistants API based ChatGPT actions stopped working. OpenAI now lists it as a past deprecation, so no extension is coming. Five actions were affected: Conversation With Assistant (Legacy), Create Assistant, Upload File with assistants, Find Assistant, and Find or Create Assistant. Zapier auto-converted only the first one, and handed the migrated Zap back switched off. Our call on the other four: three of them are deletions, not migrations, because they only ever managed assistant objects that no longer exist. Just one, Upload File, is a genuine rebuild.

What broke

What actually happened to your ChatGPT Zaps on 26 August?

They stopped. Not degraded, not warned, stopped. The deadline passed on schedule with no extension: OpenAI's deprecation list now carries the Assistants API as a past deprecation, shut down 26 August 2026, one year to the day after it was announced, and points developers at the Responses API and Conversations API instead. Zapier's own notice on the deprecation said affected Zaps would stop working after that date, and the ChatGPT actions built on the old API went with it.

Status, verified 4 September 2026 The OpenAI Assistants API is gone. It shut down on 26 August 2026, exactly one year after it was announced, with no extension and no grace period, and OpenAI now lists it under past deprecations. The replacement is the Responses API, with the Conversations API for anything that needs memory across runs.

Two things made this worse than a normal deprecation notice, and both are the reason people are finding out now rather than in August. The first is who it hit. Anyone who wired a ChatGPT assistant into a Zap during 2025 was affected, and that was the obvious way to do it at the time, so it caught the people who were early rather than the people who were careless. The second is that the notice arrived by email, roughly a year ahead, to an account owner who may not be the person who built the Zap. A year is long enough for that email to be forgotten and for the person who built the thing to have left.

What did not break is worth saying too, because the panic is usually bigger than the blast radius:

Only the steps using a retired action were affected. A five step Zap with one legacy ChatGPT action lost that step, not the Zap.
Every other ChatGPT action kept working. The newer Responses API actions, and the ordinary text actions, were untouched.
Nothing else in your account was involved. This was one integration's actions, not a Zapier platform change.

If your Zap has already broken, do these three things first In order, and none of them takes long. One: filter your Zap list by the ChatGPT (OpenAI) app and sort by status, because the workflow you are missing may not be erroring, it may be sitting switched off from Zapier's own auto-migration. Two: for anything that ran between 26 August and today, check what the downstream steps wrote, since a Zap that half-ran can leave blank rows and empty fields that look like data. Three: only then start rebuilding, using the table in section two, because three of the five actions do not need rebuilding at all.

IV Consulting take The honest framing is that this is a twenty minute job for most teams and a genuinely annoying afternoon for a few. The teams who lose a day are the ones who built an assistant with uploaded files and never documented which Zap owned it. Nothing about missing the deadline changes the work, it only changes the order: you are repairing in production instead of migrating in a window, so find out what wrote bad data before you start swapping steps.

The list

Which Zapier ChatGPT actions were retired?

Five actions, plus two dropdown fields most lists left out. Here they are with our disposition for each, which is the part the vendor notice does not give you. Zapier told you what was going away. It did not tell you that three of these are not worth rebuilding at all.

The reasoning behind that verdict is simple once you see it. Create Assistant, Find Assistant and Find or Create Assistant do not do work. They create and look up assistant objects, a concept that exists in the old Assistants API and simply has no equivalent in the Responses API. There is no replacement to migrate them to because there is nothing left for them to point at. In practice they are scaffolding wrapped around one step that produced the output your Zap actually used.

Every ChatGPT (OpenAI) action Zapier retired on 26 August 2026, from Zapier's own deprecation notice, with IV Consulting's disposition for each
Retired action	What it did	Zapier's replacement	Our call
Conversation With Assistant (Legacy)	The step that carried the real prompt and returned the output	Auto-converted to Conversation	Verify and re-enable
Upload File (with assistants)	Pushed a document into the assistant so it could answer from it	No direct replacement	Rebuild, the only real one
Create Assistant	Created an assistant object	No direct replacement	Delete
Find Assistant	Looked up an existing assistant by name or ID	No direct replacement	Delete
Find or Create Assistant	The combined lookup plus create step	No direct replacement	Delete
List Assistants (dropdown)	Populated an assistant picker inside a later field	Went away with the actions	Repoint the field
List Assistant Files (dropdown)	Populated a file picker inside a later field	Went away with the actions	Repoint the field

The detail that catches people out Zapier did auto-convert Conversation With Assistant (Legacy) to the new Conversation action, which sounds like the problem solving itself. It is not. Zapier states that migrated Zaps are turned off so you can review them. An auto-migrated Zap is a disabled Zap until someone opens it, checks that the model, prompt and output fields survived the conversion, and switches it back on. If you read the email, thought "handled", and moved on, this is the single most likely reason a workflow of yours has been quietly dead since August.

The checklist

How do you find every Zap that broke, in 20 minutes?

Do this once, properly, and you never have to think about this deadline again. The point of the audit is not to fix anything yet. It is to turn "something somewhere is broken" into a finite list you can look at. Run it even if nothing has obviously failed, because the worst case here is silent.

Filter your Zap list by the ChatGPT app

In your Zap list, filter by the ChatGPT (OpenAI) app. That gives you every Zap touching the integration, which is almost always a much shorter list than people expect. Write the count down. This is your denominator.

Open each one and read the action name

You are looking for exactly five names: Conversation With Assistant (Legacy), Create Assistant, Upload File, Find Assistant, Find or Create Assistant. Zapier flags deprecated steps in the editor, so they are visible rather than hidden. Anything else in that list is fine and needs no work.

Check the Zap history for each one too. The run that failed on or after 26 August tells you the exact date the workflow went dark, which is what you need to work out what did not get written in the meantime.

Check for Zaps that are already switched off

This is the step people skip and the one that costs them. If Zapier auto-converted a legacy conversation step, that Zap was disabled for your review and has been sitting idle ever since. Sort by status, not just by app, and look for anything off that you did not turn off yourself. On a post-deadline audit this is usually where the missing workflow turns up.

Note which Zaps involved an uploaded file

Any Zap using Upload File, or any later field populated by the List Assistant Files dropdown, is the one genuine rebuild. Flag these separately, because they need a decision about where the document lives now, not just a swapped step.

Trace what reads the output of each affected step

The failure mode that actually bites is a downstream step mapped to a field from an action you deleted. Before you remove anything, note which later steps reference its output, so you can repoint them at the replacement step rather than discovering broken mappings at run time. If the Zap has been failing for a fortnight, this is also the trace that tells you which records need backfilling.

Do this bit even if you delegate the rest Step three takes two minutes and is the only one with a silent failure mode. Everything else on this list announced itself on 26 August with a visible error. A Zap that Zapier disabled for review just stopped producing output, and nobody got an alert, which is exactly how a workflow goes missing in August and gets noticed in September. If you do nothing else today, sort your Zaps by status.

The replacement

What should you rebuild those steps with?

Zapier's notice points at two replacements, Conversation for ongoing dialogue with memory and Send Prompt for standalone requests. Its Responses API release added narrower actions alongside them: Analyze Text, Analyze Images and Extract Structured Data.

Here is the opinionated part. Most people will reach for Conversation because it sounds like the closest match to the assistant they had, and for most Zaps that is the wrong instinct. An assistant object gave you persistent memory whether or not your workflow needed it. A Zap that receives a form submission, asks ChatGPT to classify it, and writes the result to a sheet never needed memory at all. It needed one request and one answer. Rebuild that as Send Prompt and the step gets simpler, cheaper to reason about, and stops carrying state that can leak between unrelated runs.

Choosing between Zapier's Responses API actions when you rebuild a retired ChatGPT step
Pick this	When the old step was doing this	Carries
Send Prompt	One request, one answer, no reference to anything the assistant saw before. This covers most Zaps.	No memory
Conversation	Genuine back and forth across runs, or the assistant needed attached files or a live web lookup to answer.	Conversation history, file search, web search, code execution, remote MCP servers
Extract Structured Data	Prompting the assistant to return JSON so a later step could map the fields.	Structured output
Analyze Text	Summarising, classifying or scoring a chunk of text.	Single request
Analyze Images	Reading a screenshot, receipt or photo that arrived in the trigger.	Single request

The Upload File rebuild is the one that deserves a real decision rather than a swap. Your options are to move the document handling onto the Conversation action's file search, or to take it out of the Zap entirely and let the system that already stores the file answer from it. If the same handful of reference documents were being uploaded over and over so an assistant could answer questions about them, that was never really an automation step. It was a knowledge base with a Zap bolted to the front, and this deadline is a reasonable moment to say so.

IV Consulting take Rebuild to the simplest action that reproduces the output your next step consumes, not to the one that most resembles what you had. Forced migrations are the only time anyone re-reads a workflow they wrote a year ago, and the usual finding is that half the steps were there to satisfy the tool rather than the business. A broken Zap is an even better prompt for that question than a looming one, because you already know what stopped and nobody noticed. If you are weighing this against building the thing properly, our Automation stage is where that work lives.

The bigger question

Should you rebuild in Zapier, or move the workflow?

You are paying the rebuild cost either way, and now you are paying it on a workflow that is already down. That makes this the cheapest moment you will ever get to ask whether the workflow belongs where it is. That question is worth asking carefully rather than quickly: when Zendesk set a removal date on its legacy AI agents, our verdict was still to upgrade in place, because a vendor deadline is a bad trigger for a vendor decision.

Rebuild in Zapier if

The Zap is short, the trigger is an app Zapier handles natively, and your task volume is comfortable inside your plan. This is most affected Zaps. Swap the step, repoint the mapped fields, test with real data, done. Do not manufacture a migration project out of a fifteen minute fix.

Look at moving if

The forced rebuild has landed on a workflow you had already outgrown: a long multi step Zap, high task volume making the bill hurt, or logic you keep patching. That is a workflow asking to be rebuilt somewhere with real branching and version history.

Just delete it if

Nobody can name what consumes the output. Run the audit and you will find at least one Zap that has been quietly producing rows nobody reads. A deadline is a good excuse to stop paying tasks for it.

Before you pick, run your own numbers Rebuild in place or move is a cost question, not a taste question, and the honest inputs are your task volume, how many hours the workflow saves, and what rebuilding it somewhere else would take. Put those into the AI agent ROI calculator and you get a payback figure for your own workflow instead of a general opinion. Do that before you commit an afternoon to either path.

If the answer comes back "move", we have already written the version of that with the audit steps and the honest list of what does not transfer: the Zapier to n8n migration guide. And if you are weighing the platforms rather than a single workflow, Zapier Agents vs Make AI Agents vs n8n covers where each one actually holds up.

One closing thought, because it applies well beyond this deadline. Every AI feature you wire into a workflow is built on somebody's API, and APIs get deprecated. Whole products get retired too, as teams on Project Online are finding out ahead of 30 September 2026, which is the next one of these already on the calendar. The teams who handled 26 August in an afternoon were not the ones who picked the right vendor. They were the ones who knew which workflows existed, what each one fed, and who owned it. That inventory is worth more than any individual tool choice, and the cheapest time to start keeping one is right after a deadline has just proved you needed it.

FAQ

Questions teams ask now the ChatGPT Zap deadline has passed

What happened to my ChatGPT Zaps on 26 August 2026?

Any Zap step still using one of Zapier's Assistants API based ChatGPT actions stopped working on 26 August 2026. OpenAI announced the deprecation on 26 August 2025 with a one year sunset, shipped it on schedule, and now lists the Assistants API under past deprecations, so no extension is coming. Only the steps using a retired action were affected, so every other ChatGPT step and every other app in the same Zap kept running.

Which Zapier ChatGPT actions were retired?

Five actions and two dropdown fields. The actions were Conversation With Assistant (Legacy), Create Assistant, Upload File when used with assistants, Find Assistant, and Find or Create Assistant. The two dropdowns that went away are List Assistants and List Assistant Files, which matters if a later step in your Zap read a value from one of them.

Did Zapier migrate my ChatGPT Zaps automatically?

Only one of the five. Zapier automatically converted Conversation With Assistant (Legacy) to the new Conversation action built on the Responses API. The detail that catches people out is that migrated Zaps were turned off so you could review them, so an auto migrated Zap is not a working Zap until you open it, check the step and switch it back on. That is the most common reason a workflow has been quietly dead since August. The other four actions had no direct replacement and must be rebuilt or removed by hand.

What replaces Create Assistant, Find Assistant and Find or Create Assistant?

Nothing, and that is the correct outcome rather than a gap. Those three actions existed only to create and look up assistant objects, and the Responses API has no assistant objects for them to point at. In most Zaps they were setup plumbing wrapped around a single step that did the actual work. Delete them and keep the step that produces the output your later steps use.

What is the difference between the Conversation and Send Prompt actions?

Conversation keeps state across runs and carries the built in tools. Zapier lists conversation history, file search, web search, code execution and remote MCP server support on the Conversation action, so it is the replacement when your assistant genuinely needed memory or attached files. Send Prompt is a single request with no memory, which is what most Zaps actually needed. Zapier's Responses API release also added Analyze Text, Analyze Images and Extract Structured Data for narrower jobs.

Should I rebuild the workflow in Zapier or move it somewhere else?

Rebuild in Zapier if the Zap is short, runs on a trigger Zapier handles natively and the task volume is low. Look at moving it when the forced rebuild lands on a workflow you had already outgrown: long multi step Zaps, high task volume, or logic you keep patching. A forced migration is the cheapest moment you will ever get to change platform, because you are paying the rebuild cost either way.

I missed the deadline and my Zap is already broken. Can I still fix it?

Yes, and the repair is the same work it would have been in August. Nothing was deleted from your account: the Zap still exists, the step is flagged as deprecated, and you swap it for a Responses API action. What missing the deadline adds is a data question rather than a rebuild question. Find the date the Zap last ran cleanly, then check what the downstream steps did or did not write between then and now, because a Zap that fails mid-run can leave partial records that look worse than an empty one. Fix the gap first, then rebuild the step.

Who wrote this

Ishan Vats

Founder, IV Consulting · AI & automation consultant

I build automations and production AI agents for teams from startup to enterprise, across Zapier, Make and n8n. 150+ ops transformations over 10+ years. Deprecation deadlines like this one are half our inbox every quarter, which is why the advice above is mostly about deleting steps rather than rebuilding them.

Price my own rebuild →

Keep reading

Related guides and work

Migration

Zapier to n8n: the audit first playbook

If the rebuild has you reconsidering the platform, this is the honest version of that move.

Read the guide →

Relay.app shutting down, where to migrate in 2026

Migration

When your automation tool shuts down

The same forced move, run end to end after Relay.app announced it was closing.

Read the playbook →

Work with us

The Automation stage, built for you

Workflows rebuilt properly, documented, and owned by you rather than by one person's memory.

See the offer →

Rather not spend August rebuilding Zaps?

Book a free 30-minute strategy call. We will look at what the deadline actually hits in your account, tell you which steps are a delete and which are a real rebuild, and give you the plan on the spot. If it is a twenty minute job you can do yourself, we will say that and you can go do it.

Book a Free Strategy Call →

Free 30-minute call. Honest take, even if that means "you do not need us yet."