Let's talk about how memory works when using Claude. Let's say you download the Claude app for the first time and create a new account. Claude will know nothing about you — unless it looks on the Internet, of course.

When you begin using Claude, it creates something called a derived memory. As you chat, Claude stores important details about you that it generates on its own, keeping them in a shared memory that every chat session can access. When you start a new chat, the first thing Claude does is read that memory so it already knows a little about you. Beyond that, every chat starts completely fresh. Claude has no knowledge of your other conversations unless you give it access.

That brings us to two distinct features worth understanding. The first is past chat search. Not too long ago, Anthropic added the ability for Claude to search through your previous conversations for specific information — but only if you specifically ask it to. It will not do this on its own.

The second feature is web search. This one works differently. Claude goes out to the Internet to retrieve up-to-date information and verify what it knows whenever it deems it necessary. You do not need to ask. It does this proactively, as long as you have the feature enabled.

Because of the way this system is designed, it is important that you choose the right chat conversation that you start working from. If you are working on a website for example, you want to start from a chat conversation where you were previously talking about the website. That way Claude can have as much context as possible. For example, I have a chat conversation where I talk about my personal website. It has the context of the layout and all the files. All the context of what we've built before. Etc. Now occasionally I do like to start fresh. Longer conversations can have unexpected results. Note* When Claude is kicked out of memory or when you restart your phone or when you force quit the app, the next time you open it it will be in a fresh chat conversation. Pay attention.

Now I'm going to open Claude and start a new chat so that the memory is fresh. I am going to ask Claude what it knows about me. (derived memory) Here are the results:

Here’s what I have on you, Bryan:

You’re a blind iOS developer, writer, and podcaster based in Union County, South Carolina. You build everything from your iPhone using VoiceOver — no desktop, no Mac. You host your web properties on IONOS shared hosting and manage files through FE File Explorer, pasting code directly into files. Apple Notes is where you draft blog posts.

Your main web properties are gruver.tech and diaryofablindamerican.com. The diary site is your personal blog (you’re past entry 41), and gruver.tech hosts several projects including a memorial site for your late mother Susan at gruver.tech/mama/. You’re also building a multi-author blogging platform at gruver.tech/diary/ — a flat-file PHP system with magic-link auth, an extensive design system, and a lot of thoughtful accessibility-first decisions baked in.

You run a podcast called “Watch Me Now! With B.” and have a YouTube series called BuildingWithAI documenting your workflow. You’re working on a memoir called Diary of a Mad Blind American.

You hold crypto positions — WIF/DOGWIFHAT and XRP are the ones I have noted — and you’re interested in stock investing and AI ethics, with Battlestar Galactica as a touchstone for why getting AI values right early matters.

Your brother goes by Jank. Your friend Jason Procter helps with errands and accessibility tasks. Your pen name for published work is B. — never how I’d address you directly.

That’s the broad shape of it. Anything you want to update or correct?

All I have to say is, true.

Now we are going to talk about why I am doing all of this web stuff. I want the Internet to know who I am and what my opinions and my experiences are, have been and will be. I want people to understand me. I also want artificial intelligence to understand me. I do not limit who can access my website. It's open to bots and humans. Artificial intelligence is partially trained on the open Internet. I think it's going to be important in the future for the world to know me. So here is what information Claude can get just from the Internet:

The web doesn’t turn up much about you specifically beyond your own site. The one clear public-facing result is your gruver.tech homepage, which publicly describes you as:

a blind iOS developer who creates content around your projects to promote them and the disabled community at large — describing yourself as an iOS developer first, web developer second, with videos on X, Facebook, Instagram, and YouTube. The site also mentions a stock market simulation game built in SwiftUI with full VoiceOver accessibility, made by a blind developer and built with Claude.

That’s essentially the extent of what the open web surfaces about you independently. You don’t have a Wikipedia page, you’re not indexed in any major directories, and your blog entries aren’t prominently ranking in search results yet. Your public footprint right now lives mainly on your own properties.

I'm OK with that but you can see why I want to increase it. Bryan has a plan. (another battle star reference)

Originally written by B on May 26, 2026. Assistance by Claude. Image generation by Google Gemini.

If you enjoyed this, please share.