Deep Research Pro Shuts Down 23 October. A Cited Report Is Not a Signed Script
Google shuts down deep-research-pro-preview-12-2025 on 23 Oct 2026. April Deep Research agents still write cited reports, not signed explainer scripts.
Deep Research Pro Shuts Down 23 October. A Cited Report Is Not a Signed Script
23 October 2026 is the shutdown date Google prints on deep-research-pro-preview-12-2025. The Gemini API changelog (8 October 2026) says that agent is deprecated and tells you to move interactions.create to deep-research-preview-04-2026 or deep-research-max-preview-04-2026. The deprecations table lists the same shutdown, names the April speed agent as the recommended replacement, and still prints no shutdown date on either April ID. Those agents plan, search, and synthesize cited research reports. They are not a signed explainer voiceover, not a legal line, and not an 8-second clip. ClipCanva is not affiliated with Google, VEED, Kapwing, or Synthesia. Lock the claim on the AI script generator or the explainer video script generator. Pull chapters from the AI video summarizer if the source is already a recording. Steal shot language from prompt ideas. Then generate on the AI video generator or image to video. Do not brief “Deep Research migrated, so the explainer is approved.”
Product facts (confirmed 23 October 2026)
| Source | Public claim | Workflow implication |
|---|---|---|
| Gemini API release notes — 8 October 2026 | deep-research-pro-preview-12-2025 “is deprecated and will be shut down on October 23, 2026.” Migrate to deep-research-preview-04-2026 (speed / client UI stream) or deep-research-max-preview-04-2026 (maximum comprehensiveness). Update the agent parameter on interactions.create. |
A dated preview agent ID. Not a statement that Deep Research as a product is gone. |
| Gemini deprecations | Managed agents table: Pro released 11 December 2025, shutdown 23 October 2026, recommended replacement deep-research-preview-04-2026. Both April IDs released 21 April 2026 with no shutdown date announced. |
The replacement is still preview. “No date” is not “GA forever.” |
| Gemini Deep Research agent | The agent “autonomously plans, executes, and synthesizes multi-step research tasks” into “detailed, cited reports.” Preview. Interactions API only — “You cannot access it through generate_content.” Research “can take several minutes”; you must set background=true. Default tools: Google Search, URL Context, Code Execution. Optional: MCP, File Search, collaborative planning, visualization: "auto". Code samples on this page use deep-research-preview-04-2026. |
A research desk. A report with citations is not a VO card. |
| VEED AI Script Generator | Own URL. H1 “AI Video Script Generator.” FAQ still splits “Perfect your script” from “Copy script or generate video,” and links a separate AI Video tool. | Writer and renderer stay different URLs. |
| Kapwing AI Script Generator | Own URL. Separate promise to “Turn your script into a video in one workflow.” | Script generation is still a script page. |
| Synthesia Explainer Video Maker | Own URL. Flow: start with an idea → review outline → customize in the editor → generate. | An explainer maker is not interactions.create. |
Google did not publish, on the pages above, a hard clock time for the 23 October cut, a promise that April IDs write Legal copy, a ClipCanva credit map, or a proof that deep-research-pro-preview-12-2025 already returns 410. If a capability is not on the page you are briefing from today, it is not a brief. This machine did not execute a live interactions.create against the Pro ID. Treat a 404, 410, or a new banner as a new screenshot.
This is not the 19 October Notebook producer-notes note. That piece is a Review toggle on Slide Decks and Video Overviews. Today’s desk is a managed research agent on the Gemini API. Do not collapse them into “Google research.”
What 23 October shuts down — and what it does not
Operators will collapse four objects into “Deep Research died, point the explainer at Max.” The public pages refuse that.
| Surface | What the public page says it does | What it does not do | Fail condition |
|---|---|---|---|
deep-research-pro-preview-12-2025 |
Listed with shutdown 23 Oct 2026 | Prove the April IDs are GA; write a signed CTA | Leaving Pro in prod because “Deep Research still has a docs page” |
deep-research-preview-04-2026 |
Speed / efficiency; “ideal to be streamed back to a client UI”; no shutdown date announced | Sign Legal copy; become a two-minute explainer; inherit Pro’s old traces | Briefing “we migrated, ship the webinar cutdown” |
deep-research-max-preview-04-2026 |
“Maximum comprehensiveness for automated context gathering and synthesis”; no shutdown date announced | A longer report = a longer voiceover; a chart = a locked still | Picking Max because the nickname sounds like the campaign |
Interactions API + background=true |
Required path. Not generate_content. Minutes, not a one-shot chat |
A YouTube disclosure; a 9:16 crop; a SKU still | Calling generateContent and calling the 410 “an outage” |
| VEED / Kapwing script URLs | Generate a script | Disclose on YouTube; pick 6s vs 60s; lock a SKU still | Hitting generate-video from an unsigned research memo |
| Synthesia explainer URL | Idea → outline → editor → video | Grep your repo for -pro-preview-12-2025 |
Treating avatar output as the research citation |
If you are calling deep-research-pro-preview-12-2025, you are on a dated Gemini API preview desk. If you are calling either April ID, you are on a still-preview research desk whose own shutdown line is not announced. If the bottle still needs a legal sentence, you are on a writer. If you cannot tell which, do not cut over.
Pick the desk before you rewrite the agent string
The fail is treating “shutdown: 23 October” as “the explainer is covered.” Map the job.
| Job | Where it actually runs | What you must supply | What the system may invent | Fail condition |
|---|---|---|---|---|
| One claim the clip must carry | AI script generator or explainer video script generator — VEED / Kapwing if that is already your writer | Audience, seconds, the signed sentence | A second CTA, a health claim, a competitor name | Skipping the card because “the report is cited” |
| Recap of a long cut or webinar | AI video summarizer | Source cut, claims to keep, claims to strike | Compression that changes meaning | Feeding the PDF into Deep Research and calling the summary a master |
| Motion from a locked still | Image to video | The signed still as frame one | Camera path, extra hands, a spoken line you did not approve | Animating a chart the agent drew because visualization: "auto" was on |
| New motion without a locked frame | AI video generator or a named model on Compare | Locked caption, crop, seconds | Props, voices, a second location | Briefing Deep Research Max as a two-minute explainer renderer |
| Model / agent pick | Compare plus the live changelog and deprecations table | Job = Pro vs April-speed vs April-max vs a non-Google writer | A Notebook screenshot from last week | Picking “Deep Research” as if it were one string |
VEED still keeps AI Script Generator as its own URL and AI Video as another. Kapwing still keeps a script generator and a “turn your script into a video” promise as different beats. Synthesia still keeps an explainer URL with an outline review before generate. None of those pages say a Google agent-ID swap writes your CTA.
If the brief is “grep the repo for deep-research-pro-preview-12-2025,” that is engineering. If the brief is “the two-pack restocks Friday on YouTube,” that is script, still, then image-to-video. If the brief is “Max will handle the clip because comprehensiveness is higher,” you do not have a brief.
Creator / operator checklist
- Name the string out loud.
deep-research-pro-preview-12-2025≠deep-research-preview-04-2026≠deep-research-max-preview-04-2026. Screenshot theagentfield in logs, not the nickname on a slide. - Grep before you generate. Search production, staging, notebooks, and vendor dashboards for
deep-research-pro-preview-12-2025andpro-preview-12-2025. A hidden retry queue is still Pro. - Do not treat 23 October as “Deep Research is dead.” The docs page still exists. The April IDs have no shutdown date announced. That is a different calendar from Pro.
- Do not treat April as GA. Both replacements are still labeled preview on the Deep Research page. Preview is a stage, not a brand.
- Stay on Interactions. The docs page says you cannot call this agent through
generate_content. A 400 on generateContent is not evidence the shutdown “already happened.” - Keep
background=true. The same page says research can take several minutes and you must run it asynchronously. A frontend timeout is not a model outage. - Cap the spoken line before you cut over. Use the AI script generator or the explainer video script generator. One sentence. Strike “#1,” ingredient claims, and competitor names you cannot defend. A cited report will happily imply a claim Legal did not sign.
- Do not ship agent charts as pack shots.
visualization: "auto"can emit charts when the prompt asks. That image is research support, not a locked SKU still. If the bottle must survive, lock frame one on image to video. - Keep writer, still, and renderer as three objects. A research memo, a signed line, a locked frame, and a YouTube disclosure are four things. Do not hide all four under “we swapped Pro for Max.”
- Re-screenshot changelog + deprecations on cutover day. Changelog entry is 8 Oct. Deprecations table still prints 23 Oct 2026 for Pro. If either banner changed after you read this, the ticket follows the new screenshot.
Worked example: a Friday explainer that does not start in deep-research-pro-preview-12-2025
Say the job is a YouTube explainer cutdown. The two-pack ships Friday. Legal line approved: Same bottle. Same seal. Restock Friday. Someone opened the 8 October changelog, saw 23 October, saw Max, and told the channel “Pro is shutting down, so we point the job at deep-research-max-preview-04-2026 and the explainer is covered.”
That sentence mixes five desks.
The spoken line is a writer object. Put the approved sentence on the AI script generator or the explainer video script generator. Deep Research can return a cited report — the docs page says that is the job. It will not know which sentence Legal signed unless you paste that sentence. If you already live in VEED or Kapwing, their script tools are still script tools. Copy the signed line. Do not hit generate-video from the unsigned memo.
The beat list is a card object. Steal shot language from prompt ideas, then write four lines, not a chat novel:
- 0s — kitchen, bottle already on the board (do not invent a second SKU)
- 2s — hand in, label readable
- 4s — pour or use, same bottle
- 6–8s — hold the signed line, no new claim
If the source is a 40-minute webinar, run it through the AI video summarizer after you mark claims to keep. Do not paste the webinar into Deep Research and call the report a VO.
The picture is a still object. If the label must match the listing, photograph or render the approved pack, then animate that file on image to video. If visualization: "auto" drew a market-share chart, that chart is not the bottle.
The motion is a renderer object. Use the AI video generator or a named row on Compare. Deep Research Max is still a research agent. Synthesia’s explainer URL still asks you to review an outline before generate. Neither page is a stay of execution for Pro.
Engineering still has a ticket: grep for deep-research-pro-preview-12-2025, move interactions.create to an April ID if the job is still research, keep background=true, and screenshot the deprecations row. That ticket does not ship Friday’s cut.
FAQ
Did Google shut down Deep Research today?
Google’s changelog and deprecations table shut down one agent ID: deep-research-pro-preview-12-2025, on 23 October 2026. The Deep Research docs page still describes the product, in preview, on the Interactions API. The April IDs have no shutdown date announced.
Is deep-research-max-preview-04-2026 the “full” replacement?
The deprecations table’s recommended replacement for Pro is deep-research-preview-04-2026. Max is the other April option, described as maximum comprehensiveness. Neither page says Max writes a signed explainer.
Can I call Deep Research with generate_content?
The Deep Research page says no: “You cannot access it through generate_content.” Use interactions.create with background=true.
Does a cited report replace a script? No. Citations are a research property. A YouTube line is a writer property. Lock it on the video script generator or the explainer script page, then render.
Should I wait to see if Pro already 404s? If production still sends Pro, grep it today. This article did not probe the live endpoint. A new status code is a new screenshot, not a reason to skip the writer desk.