
WorkBuddy and WorkGPT aren't the same kind of thing — I ran one translation job through both
Let me put the conclusion up front. A plugin like WorkGPT stuffs a sidebar into every Google app — one in Docs, one in Sheets, one in Gmail — each doing its own thing. WorkBuddy is a workbench: you lay out files, fields, permissions, and it runs the whole thing in one pass. For work like free translation, pick the latter.
No gigs today, so I scrolled onto WorkGPT's app detail page, updated Nov 3, saying it supports Docs, Sheets, Slides, Gmail, Calendar, and the model can switch among GPT-5, Gemini, Claude, DeepSeek and that whole lineup. Looks comprehensive. But I didn't rush to install it — I thought of last week's job first.
Client sent twelve PDFs, two glossaries, and wanted a unified-format delivery sheet plus a comparison table. If I split this across those Google apps, I'd have to create a folder in Drive, then open each one, copy, paste, align fields. WorkGPT can rewrite a paragraph within a single doc, but it can't produce results from twelve files against one set of fields.
I've used WorkBuddy for three months, so I know this path well.
Open WorkBuddy, the task area is on the left, click New Task, choose File Processing. First-timers tend to make a mistake here — they just click upload. Hold on. First define the output fields.
Mine goes like this: after the task is created, go into Output Settings and manually add columns. Source filename, page count, word count, glossary hits, delivery status, notes. Six columns, no more no less. This step is key — once the fields are set, it knows where to fill. Two months ago when I first started using CSV I didn't get this, dumped all the PDFs in at once, and the resulting table had garbage column names — I had to fix it myself for half an hour.
Once fields are set, go back up a level and drag all twelve PDFs into the Input Area at once. It queues them up and goes through one by one. Now go to the Glossary field and import the CSV I prepared — two columns, source on the left, translation on the right, UTF-8 encoding. I've used glossaries for two months; the pitfall I hit is encoding — a CSV saved directly from Excel carries a BOM, so when imported the first column name has an invisible character in front of it and all matching fails. When saving, choose "CSV UTF-8," or just run it through a text editor.
Once running it takes a few minutes. On the right side of the interface is the task log, scrolling down line by line, showing which file finished parsing and which got stuck. Two scanned files were photographed crooked and it flagged them "low confidence" — I clicked in and fixed two spots by hand. This is the pitfall-avoidance I mean: don't expect full automation; leave a manual review pass, it's faster than rerunning.
The results come out in the Results Area, one table, six columns filled. Click Export, choose PDF. For the cover image field I left it blank — I got burned last time, the image got stretched out of shape, and I later found it was because the prompt didn't hard-code the aspect ratio. This time I turned off the cover image in Page Settings, pure table, clean.
One word on permissions. WorkBuddy's collaboration isn't dumping the whole task on someone else. Click Share in the top right, choose View Results Only, generate a link and send it to the client. The client can see the exported table, can't edit the source files, can't get into task settings. My finalized files go in the Archived section, locked, only I can unlock. I used to think this feature was redundant; now I've changed my mind — client fumbles and changes one glossary term, and you can't even find which version it was.
Daily upkeep is just two things. Clear History Tasks once a week — archive or delete anything over thirty days, otherwise the list keeps growing and it's a pain to find things. Save a copy of the template locally — field config and permission settings are all in it — and next time you create a task just apply it, saving the ten minutes of reconfiguring.
Back to WorkGPT. Its positioning is different. Its strength is in-place rewriting, generating slides, summarizing emails within the Google ecosystem, switching models freely. But ask it to run twelve PDFs against one set of fields — it has no concept of that. It has no task area, output settings, glossary import, and no "who can view, who can edit" layer.
My guess is that over the next year these tools will split into two camps. One camp lives inside apps, doing single-point assistance; the other is a workbench, managing process and delivery. In translation, the second will become the default setup. After all, what the client wants is something they can check against — speed alone is useless.
Physix Frontier