Your best ideas are already
in your old chats.

Your best ideas are already in your old chats

You've done more real work in AI chats than you remember. Most of it has scrolled out of sight.

On October 10 I released History Extractor in the Skills Builder Library. You type one sentence, and it goes back through your past Claude or ChatGPT chats. It finds the ones where you made decisions, wrote in your own words, heard from customers, or started something you never finished. Then it hands you a dashboard you can work from the same day, plus a plain copy to keep.

I built it because of what happens to me most. I open a chat to do one thing, the AI hands me a recommendation, I follow it, and by the end of the thread the idea I came in with is buried way up the scroll.

Six ways it's useful (pick the one that sounds like you)

• A long thread wandered off. Run it on that one chat. It finds what you came in to do, shows where the thread turned and whether that turn was a fix you needed or a shiny object, and gives you the next step back.
• You keep re-explaining your business. It writes one snapshot paragraph you can paste at the top of any new chat or into your project instructions.
• You're about to write something that sells. It pulls lines you actually wrote and what customers actually told you, so your sales page or launch emails start from your words instead of AI's.
• You're setting up your context files. It drafts copy-ready updates for who-i-am, my-business and my-voice from your real history instead of a blank page. It never changes your files for you.
• You started something good and lost track of it. Its “lost gold” section lists up to three promising things that went quiet, where each one stopped, and the way back in.
• You just came out of a busy stretch. “Start here” gives you up to three things worth doing this week, each with one concrete action.

It isn't only for business, either. Point it at a long chat about your day job, a decision at home or something you're teaching yourself, and the dashboard is built around what you were trying to solve there.

What it looked like for a member

One member ran it on October 13. The ideas it surfaced matched the parked-ideas list she'd already pulled together on her own. That's the check you want: it found what was really there. And before you reuse anything, the dashboard gives you a short list of prices, statuses and quotes to confirm first.

Don't have the Skill? Try the one-thread version

Open the long chat that wandered off, scroll to the bottom, and paste this. You don't need to fill anything in.

What this does and doesn't do: it works on the one chat you paste it into, in Claude, ChatGPT or Gemini. It won't search your other chats, find your lost gold, pull out your own words, build the dashboard, or draft updates for your context files. That's what the Skill is for.

History Extractor is included in the Skills Builder Library for VIP and Founding Premium members.

Why this is the right job for AI

Digiday talked to creators this month who've landed in a similar place: they use AI for their inbox, their calendar and their files, while some brands are writing “no AI” clauses into campaign briefs and creators are noticing that a lot of AI writing has started to sound the same. Digging through months of old chats is back-office work. AI can do the digging. The judgment about what's worth keeping, and the voice you write it in, stays yours.

ChatGPT got a new brain, and it might show you charts now

Everyone's ChatGPT changed on October 7 and 8, whether you noticed or not.

GPT-6 now runs regular ChatGPT conversations. On Plus and the plans above it, that's GPT-6 Sol. Free and Go got the upgrade a day later. In the model menu, GPT-6 replaces what used to say Latest. OpenAI's model guide

The visible change is something OpenAI calls Intelligent UI: answers can now include charts, buttons and small interactive tools, and it can start answering while it's still thinking. Release notes

What to do: run one task you do every week and compare. If the visuals help, keep them. If they're noise, go to Settings and turn down Layout and visuals. Older desktop app versions don't show the new visuals, so update the app if you want them.

Why it matters: this is the first change in a while that every ChatGPT user got automatically. It's also why some answers may suddenly look different from what you're used to.

Heads-up on naming: GPT-6 Sol (regular Chat) and GPT-6.1 Sol (ChatGPT Work and Codex only) are different things. If you don't use Work or Codex, you don't need 6.1 Sol. GPT-5.5 was retired from ChatGPT on October 14, which only matters if you'd picked it on purpose.

Claude moved into your documents

Two changes, one direction: less copying and pasting.

• Claude inside Google Docs, Sheets and Slides (October 6). A Claude sidebar works on the document you already have open. It's in public beta on all paid Claude plans, and it can only reach what your Google sharing permissions already allow.
• Claude's own Docs, Slides and Design are out of beta on every plan, Free included (October 8). If you've been on the free plan, you can now make a document or a slide deck right in Claude. On paid plans there's also a new Dashboards beta that turns a spreadsheet into a live view you can ask questions about.

What to do: if you pay for Claude and live in Google Docs, try the sidebar on one real document this week. Something you'd normally copy into a chat and paste back is a good test. If you're on Claude Free, make one document or deck from notes you already have.

Why it matters: the work moves to where your files already are. That's less copying back and forth, and fewer versions to keep track of.

Two deadlines worth five minutes

Only matters if you built on these. If you didn't, skip this section.

• Custom GPTs retire December 11. If you built one, migrate it or back it up now. When you migrate, your instructions become a skill and your knowledge files come along. The model you picked, custom actions and sharing don't carry over, and unpublished drafts don't transfer. So publish any draft you want to keep first. Most personal plans already can't create new GPTs. If you only use GPTs other people made, there's nothing to do yet. OpenAI's FAQ
• Google Opal and the Gems made by Google Labs shut down November 17. Your files stay in the “Opal” folder in your Google Drive, but the workflows stop running. Before then, open each one and copy any prompts or instructions you want to keep. Opal FAQ Separately, your regular Gemini Gems are becoming “skills.” For personal accounts that happens in November, and Google says Gems move over automatically with their knowledge files. Workspace accounts have until March 2027, and Google has delayed that rollout. Gems FAQ

Short notes

• ChatGPT Slides (October 16): ask ChatGPT in Work to make a presentation from a chat, your files or an outline, or share an example deck so it copies the style. You can edit it yourself or ask ChatGPT to, present from ChatGPT, or export to PowerPoint, PDF or Google Slides. It's rolling out to Pro, Business and Enterprise with Space access. Plus and Free aren't on the list yet, so if you're on Plus there's nothing to do. OpenAI's Slides guide
• ChatGPT Voice got a new model and new voices (October 15). Live voice now runs on GPT-Live-1.1, with more natural accents and new voices in several languages. Use is counted over a rolling 24 hours: 3 hours on Plus, 15 on the $100 Pro plan and unlimited on the $200 one, while Free and Go get a smaller version. If you think out loud, try talking a draft through before you write it. OpenAI's Voice guide
• ChatGPT Sites has no cap on how many sites you make (October 15, Plus and Pro). Usage and storage limits still apply. Release notes
• Gemini 4 Argon is in the Gemini app for Google AI Pro and Ultra (October 13). More in Model watch. Gemini release notes
• Gemini's free plan now defaults to Auto (October 9): most prompts go to a lighter model, and harder ones can still go to bigger ones. Gemini also has a new Customize menu where skills and connected apps live. Google's help page
• ChatGPT takes audio files now (October 6, paid plans, up to 512 MB). Drop in a voice memo or a recorded call and ask for a summary and a follow-up draft. Speaker labels can be off, and one outside test didn't get a transcript, so try a short recording first. OpenAI's help page
• Notion 3.8: if you're on a Notion Business plan and pay for ChatGPT Plus or Pro, your ChatGPT plan can now cover your AI use in Notion. Check the limits before you drop anything. Notion's release
• ChatGPT's document scanner (October 1, iPhone): photograph handwritten notes and work from them as text.
• ChatGPT Dots, a week with mine: I set mine up on October 10 and named it Evy. My first project is still waiting on me to answer a question, but that doesn't mean it hasn't finished anything. We went through my subscriptions, figured out what I was actually using, and found overlaps. I canceled seven paid subscriptions or add-ons, adding up to roughly $1,700 a year in avoided renewals. It also pulled payment dates from my email, checked them against receipts and screenshots, and caught something I needed to handle. We even worked through grocery shopping and family meal ideas. Along the way, it's surfaced time-sensitive things from my email and Notion that I might have missed. The useful part hasn't been “AI runs my life.” It's been having help turning scattered information into a clear next step. I'm still making the decisions and taking the actions, but I'm doing less digging to figure out what needs my attention. Dots are on Pro (not in the UK, EEA or Switzerland) and Business Premium, and Enterprise admins can turn on a beta. They're rolling out gradually, so check whether your plan has access. OpenAI's Dots guide
• Pro 200 subscribers: the earlier usage allowance ends after October 29. The price stays $200. Check your usage before then. OpenAI's Pro page

Nobody needs to switch plans this round. Two things are worth trying.

• Google AI Pro? Try Gemini 4 Argon for research. It's #1 on the Vals Index and has the lowest rate of made-up answers among the top models on Artificial Analysis. It's weaker on finished work like decks and long documents, and one reviewer calls the app around it dated. Use it to find and check facts, then click every link.
• Claude Free is the strongest free plan right now. Anthropic says anyone can chat with Sonnet 5.5 on Claude.ai, and it scores close to Opus 5.5 on work tasks. If you're on Claude Pro, use Sonnet 5.5 for everyday writing and email, and save Opus 5.5 for long projects. It's the overall leader on independent tests.
• On ChatGPT Pro? GPT-6 Pro got faster (October 16), and it can now split a big task across several agents working at once, with their progress shown in the chat. Save it for multi-part jobs, not quick rewrites. If you're not on Pro, skip this. Release notes
• Haiku 5.5 is the new fast one in Claude. It's in the model menu on every plan. It's good for quick, simple jobs. Don't expect it to do your heavy thinking.

Which model for which job

• Writing in your voice: Opus 5.5 on Claude Pro, with two or three of your own samples. Good enough and cheaper: Sonnet 5.5 on Claude Free.
• Client emails and quick replies: Sonnet 5.5, or GPT-6 in ChatGPT. Any free plan you already use works too.
• Research with sources: Gemini 4 Argon on Google AI Pro, then check every link. Cheaper: ChatGPT or Gemini with search turned on.
• Long projects and deliverables: Opus 5.5, and say what “done” looks like. Cheaper: Sonnet 5.5 for drafts, Opus for the final pass.

One tip that saves usage: don't set thinking or effort to max for normal work. The newest Claude models can overthink at max and burn through your limit. The default setting is the right one for most jobs. And if you have an unused one-time usage reset in Claude, check Settings > Usage for when it expires. Claude's help page

Safe to ignore

• Claude's monthly API credits for Max and Team: for people building their own apps. Pro isn't eligible.
• ChatGPT ads: labeled ads are coming to image generation on Free and Go, and businesses can now buy ChatGPT ads. Watch it; don't buy anything yet.
• Anthropic's Usage Policy update (November 12): policy, not a change to how you work.
• Safety research headlines: the alarming-sounding lab reports from October 9 and 12 are research warnings, not product problems.
• Builder releases: Codex betas, Grok Voice 3.0, Meta's Muse Wonder and the Gemini coding agent on Mac.