Our AI rankings hold six companies, and the spread is the widest on the site: Lockeddoor.ai at 97, ChatGPT at 90, Google Gemini at 89, Claude at 88, Character.AI at 80, Replika at 78. Nineteen points separate the best from the worst, and the gap is not about raw capability. ChatGPT writes, codes, browses, and generates images, and its review calls it the benchmark — yet it sits seven points behind the leader. The difference is memory. Ask what each of the six actually remembers about you between conversations, and the ranking stops looking like a mystery: the product is whether it knows you when you come back. Every number below comes from our review of the company, which comes from the company’s own published pages.
The one that remembers you
Lockeddoor.ai, at 97, is the #1 Top Rated entry in the category, and its verdict is the thesis of this post: “The AI companion that actually remembers you.” Its review leads with persistent memory: companions remember you, your life, and your past conversations across sessions. The relationship continues where it left off, every time. Memory is the reason the score is 97. The rest of the review backs it up: a full town of distinct characters with their own personalities and relationships, voice notes, proactive check-ins, and photo drops — the companion reaches out like a partner, not a tool waiting for a prompt. The price is one honest number: a single $9.99/mo tier and a $0.99 trial for 100 messages. Its skip line is honest: people shopping for a work assistant should look at the general-purpose assistants below. Memory has a job here — continuity in a relationship — and the review scores it as the whole product.
The assistants that remember your files
ChatGPT at 90, Gemini at 89, and Claude at 88 sit within two points of each other, and all three remember the same way: they remember your work, not you. Gemini’s review is clearest about the distinction. With your permission it connects to Gmail, Docs, Drive, Maps, and YouTube, so it can answer about your own email, files, and trips instead of starting from zero. That is real recall — of your documents. Its skip line says it plainly: users who want a personality that remembers them as a person should look elsewhere. ChatGPT’s review names the same limit: its memory is “limited and oriented to tasks, not relationships,” and its verdict is “capable at almost everything, personal at very little.” Plus is $20/mo — and the extra money buys capability, not closeness. Claude, at 88, does not even pretend: its Projects feature keeps related work organized — memory as filing — and its cons list is blunt — “Assistant-first: no companion or character dimension.” Pro is $20/mo, and its skip line matches: users who want personality or a companion relationship should look elsewhere. Three capable assistants — and a shared, published admission that none of them is trying to remember you as a person.
The companions that forget
This is where it gets uncomfortable: the bottom two sell companionship and both, on their own reviews, forget. Character.AI’s verdict is “fun first, continuity second,” and its cons list says it outright: “Memory across chats is weak; characters forget and drift.” Millions of user-created characters, free to start — and the character you talked to yesterday may not remember the conversation. Its skip line is the giveaway: anyone who wants one companion who actually remembers them over time should look elsewhere. Replika is the sadder case, because it invented the category. Launched in 2017, it keeps a diary of “memories.” But its review says the category moved: newer companions built their reputations on long-term memory, and Replika’s recall is noticeably shallower. Its cons list is just as direct: “Forgets details and loops in conversation more than top rivals,” with much of the personality behind the Pro paywall. The free tier gets you basic chat; remembering you costs extra, and even then it forgets.
What memory actually costs
Line the published prices up and the pattern sharpens. The best memory in the category is also the cheapest named price on the page: Lockeddoor.ai at $9.99/mo with a $0.99 trial. ChatGPT Plus and Claude Pro are both $20/mo — double — and both buy work capability, not memory. Gemini’s paid plan is bundled with Google One storage; it buys a better assistant, not a better relationship. Character.AI is free to start with a paid c.ai+ tier, and Replika is free with a Pro subscription for voice calls and relationship modes. So the two cheapest ways in are the two weakest memories on the page, and the two most expensive assistants remember your files best and you least. The only entry where price and memory point the same way is the one at the top — the ranking, stated as a price list.
What to watch
Three things, all checkable from the published pages. First, whether ChatGPT, Gemini, or Claude ever publish a memory feature aimed at the person rather than the task — a saved sense of who you are, not just a better search of your inbox. Any one of them could close half the gap with a single feature. Second, whether Character.AI closes its continuity gap: a library that size with real memory would be the biggest score move in the bottom half. Third, whether Replika’s Pro memory ever catches the leaders it inspired, or the pioneer settles into being the friendly first step its review describes. Until one of those happens, the category’s lesson stands as published: in AI, memory is not a feature. It is the product.
These scores are a snapshot of what the companies publish today, checked November 20, 2026. When a company changes what it publishes, its score moves with the evidence. The full AI rankings are the place to watch it change.
← Back to the blog · November 20, 2026