ChatGPT says it will write in your voice

Click the link below and a popup opens.

It says ChatGPT will write in your voice.

That is a big claim, and it is sitting on their own screen. I want to know if it is true.

Here is what they claim it does. You connect Gmail, Slack, and your documents. It reads what you have already sent and already written, and it drafts from that.

Gmail for your emails. Slack for your messages. Your documents for the long writing.

I set it up. I have not run it on real work yet, so I cannot tell you how well it holds.

Finding it was the slow part. I looked in settings. I looked in plugins. I looked in skills. It is not in any of them.

The link is the only way I got in. Open it on a computer, not your phone, in the browser where you are already signed in to ChatGPT.

It runs on the paid ChatGPT Work plans.

GPT 6 Astra sounds promising to me. That is a feeling so far, not a test.

So this is the week. Connect your apps, give it something real to write, and tell me what came back and whether it sounded like you.

I am running it on my side at the same time. Yours and mine both go in a coming issue.

Fable 5.1 came out on 1 September. GPT 6 Astra came out on 4 September.

From my own use so far, Fable 5.1 is the better writer. Astra got much better at using a computer.

Three days is a small gap, and Astra took the whole week. I think the timing was deliberate, and it worked. The attention went one way.

Then Zachi asked GPT 6 Astra to draw him in Canva.

Astra opened Canva, picked the brushes, and built his face block by block. (17.3k likes · 913 reposts · 4k bookmarks · 1.9 million views.)

People ran the same instruction on Fable 5.1, and you can see those results in the same thread. They were weak.

Drawing the way a person draws, this is the first time I have seen an AI get close. That is what people have been doing with Astra all week, their own photos, the same instruction.

That one happens inside Canva. Google shipped its own version of Canva on the same day Fable landed.

Image editing, sitting inside Docs and Slides.

Click a picture on your slide, change it there, stay on the page. It also translates the text inside an image.

I only found out about this today, so I have not touched it yet.

It is still rolling out. Open it and see whether you have it.

ChatGPT wrote a short bio for me last week and it gave me the wrong job.

Then I opened the settings. There was a summary in there, about me, written by ChatGPT, and I had never read it.

One line described work I stopped doing months ago. So I had been correcting the answer while the cause sat 3 menus away.

This is why people do not trust AI memory. You never wrote it. It built up on its own, out of old chats you have already forgotten.

So you keep typing who you are at the start of every chat, every time.

Reading it once fixes that. The memory is useful the moment you have seen it.

Every assistant keeps one, and each one puts it in a different menu.

ChatGPT: Settings, then Personalization, then Memory, then Manage.

Claude: Settings, then Memory.

Gemini: Settings, then Personal Intelligence.

Before you open any of them, paste this into the one you use.

"Based only on what you have saved about me, tell me what you think I do for a living."

If the answer feels wrong, ask it what it remembered. Then go and look.

Delete the lines that are out of date. An old job. A finished client. A city you left.

I also made a clean Claude account and gave it 7 facts about a made-up bakery. It kept 6. The one it dropped was a walnut allergy, and it did not say so.

There is a second switch under the main one that keeps health out of the list, and it is off.

One date. Claude retires its old memory summary on 9 September, which is tomorrow. You can export it until then.

Reading all 3 screens took me 5 minutes. Cleaning them took 15 more, because I stopped and read every line.

I opened all 3 screens by hand on 7 September, on free accounts, and kept the screenshots. 7 pages.

Free. No signup.