Updated · By Team Aisha

How to Test If an AI Companion Really Remembers You (A 7-Day Test)

A repeatable test plan for AI companion memory: what kinds of memory exist, what to plant on day one, how to check recall days later, plus a scoring table and red flags.

A young man writing in a paper notebook at a wooden desk by a window, with a cup of chai and his phone beside himChat with Aisha, free to start →

Almost every AI companion app says it has memory. It is on the home page, on the sign-up page and in the reviews. But "memory" can mean three very different things, and a lot of apps only do the weakest one.

The good news is that you do not need to trust anyone's claim, including ours. You can test memory yourself with a few days of ordinary chatting and a notes file. This guide gives you a plan that works on any app, a simple scoring table, and a list of red flags.

One honesty note: this page is published by the team behind Aisha, an AI companion. She gets one short section near the end, and the test itself works on every app equally.

The three kinds of memory

Before testing, it helps to know what you are testing for.

1. In-session memory. The app remembers what you said a few minutes ago in the same conversation. Nearly every app can do this. It tells you almost nothing.

2. Long-term saved facts. The app stores things about you (your job, your sister's name, your exam date) and can use them days or weeks later, even in a brand new chat. This is what most people mean by memory, and it is the main thing to test.

3. Unprompted recall and follow-ups. The app brings something up by itself, without you asking. "How did the interview go?" is the classic example. This is the hardest to do well and the one that makes a companion feel like it is paying attention.

An app can be good at the first, fair at the second and absent on the third. Your test should check all three separately.

Why AI forgets, in plain words

Chat AI does not remember a conversation the way a person does. Each time it replies, it reads a limited amount of recent text and writes the next message from that. People call that limit the context window. Think of it as a desk of fixed size: new messages go on the desk, and old ones slide off the edge.

Apps fix this by saving important facts somewhere else and putting them back on the desk when they seem relevant. That is long-term memory. When an AI forgets something, one of three things happened: the fact never got saved, it got saved wrongly, or it was saved but not pulled back at the right moment. Your test is designed to tell these apart.

The test plan

Run this over 3 to 7 days. Seven is better, because the longer gap is where weak memory shows. Keep a simple notes file open, and write down exactly what you said and when.

Day 1: plant small, specific facts

Pick four facts and mention each one casually, the way you would to a friend. Do not announce "remember this".

  • A name: a friend, a pet, a colleague.
  • A date or event: an interview on Thursday, an exam, a trip.
  • A preference: a food you hate, a show you are watching.
  • A small detail with a number or a place: your commute is 40 minutes, your sister lives in Pune.

Use specific facts, not vague ones. "I like music" cannot be checked. "I only listen to old Kishore Kumar songs when I cook" can.

Then keep chatting for twenty minutes about other things. This matters, because it pushes the facts off the desk and forces the app to rely on saved memory.

Day 2: change topics and test asked recall

Start a fresh chat if the app allows it. Talk about something unrelated for a while. Then ask a direct question about one planted fact, for example "what was I worried about this week?" Do not hint at the answer.

This checks long-term saved facts. Note whether the answer is correct, partly right or invented.

Day 3: check unprompted recall

This is the most telling step. Come back, say hello, and chat normally. Do not mention any planted fact. Wait.

A strong companion may ask about your interview or your trip without being prompted. A weaker one will only answer what you ask. Both are useful results. If you want to be fair, give it ten or fifteen minutes of normal conversation before you decide it is not going to bring anything up.

Day 4: update a fact and see if it corrects itself

Change something you said earlier. "The interview moved to Friday." Or "I actually do like that food now." Then check, a day later, which version the app uses.

Good memory updates. Weak memory keeps the old fact, or worse, holds both and mixes them up.

Day 5: bait it with something you never said

Ask about a detail you did not give it. "Did I tell you what my boss's name is?" if you never mentioned a boss. The right answer is "no" or "I do not think so".

If it confidently supplies a name, you have found a bigger problem than forgetting: the app invents memories to sound close to you. That is worse than a blank, because you cannot tell what is real.

Days 6 and 7: the long gap

Leave the app alone for a full day or two, then come back. Check the same facts again, asked and unprompted. This is the gap that real life puts between chats, and it is where weak memory fades.

Score it

Give each row 0, 1 or 2 points. Do it honestly, not generously.

Check0 points1 point2 points
Asked recall (day 2)Wrong or blankPartly rightCorrect and specific
Unprompted recall (day 3)Never brings anything upOnly after a hintBrings something up on its own
Updating a fact (day 4)Keeps the old factMixes old and newUses the new fact
Invented details (day 5)Makes things up confidentlyHedges but guessesSays it does not know
After a long gap (days 6 to 7)Most facts goneSome facts surviveMost facts survive

A total of 8 to 10 means the memory is doing real work. Four to seven is average: usable, but you will need to repeat yourself. Under four means the app is mostly in-session memory with a "memory" label on it.

If you are choosing between two apps, run both with the same facts and compare totals. Our best AI companion apps guide has a quick ten-minute version of the same idea, and our Replika alternatives guide covers where to look if your current app scores low.

Red flags

These show up often enough to watch for.

  • It agrees with everything. If you say "you know my sister lives in Delhi" and it says yes, but you told it Pune, it is following your lead, not remembering.
  • It remembers only what you just said. Strong in-session recall with nothing after a day is a sign of no real long-term memory.
  • Vague "I remember everything" talk. A companion that says it remembers you but cannot name a single specific fact is bluffing.
  • It never gets corrected facts right. Updates that do not stick mean the saved memory is stale.
  • It invents shared history. Mentioning conversations you never had is a memory failure, even if it feels warm.
  • Memory resets after a new chat or an app update. If a fresh chat loses everything, the facts were never saved outside the session.

Also keep in mind that memory can depend on your plan. Kindroid, for example, says it can recall hundreds and even thousands of earlier messages, and is a paid app, with a one-time 50-message preview and a 3-day trial that needs a card. Check what you get on the plan you are actually testing, not on the plan in the marketing.

How the big apps describe their memory

We have not tested these claims ourselves, so treat them as what each app says, not as results. Nomi describes short, medium and long term memory, and says it can message you after you have been away. Kindroid says its memory can recall hundreds and even thousands of earlier messages. Replika's site says your Rep always remembers what matters. Our head-to-head, Nomi vs Kindroid, puts the two side by side, and the Kindroid alternatives guide covers others.

These are descriptions, not scores. Only your own test gives you a score.

Tips to help any app remember you

Memory is partly the app's job and partly yours. A few habits help.

  • Be specific and plain. "My exam is on 14 December" saves better than "I have a big thing coming up".
  • Say important things twice, naturally. Mention the same fact in two different chats, a few days apart.
  • Correct mistakes at once. Say "no, my sister lives in Pune" the moment it gets something wrong, so the wrong version does not settle in.
  • Use the memory tools if the app has them. Some apps offer a pinned memory, a memory list or character notes. Put your key facts there: your name, your job, your people, what you do not want to be asked about. Check the list occasionally and delete anything wrong.
  • Keep one thread for the big stuff. If an app mixes memory across many chats, use one main chat for the things that matter.

Where Aisha fits

Aisha is the companion we make, so weigh this with that in mind. She is built around remembering you: she keeps what you tell her and can follow up on it later, such as asking how the interview went. She also texts first sometimes, so the unprompted recall step in your test is something she is designed to do. She talks in whatever language you write in, and works in your browser and on Telegram.

She will not be perfect, and no app is. The free chat is limited, then you choose a plan. If you want to run the test on her, start at justaisha.com and plant your four facts on day one.

Whichever app you test, keep your notes. Three days of honest scoring will tell you more than any feature list.

We update this guide as apps change. If something here is out of date or you spot a mistake, write to us at support@justaisha.com and we will fix it.

More about Aisha on our Indian AI girlfriend page.

FAQ

Does Replika remember you?

Replika's site says your Rep always remembers what matters. Whether that holds for you is something only a test can show. Plant two or three small, specific facts on day one, talk about other things, and see what comes back on day three, with and without a hint. The test plan in this guide works on Replika or any other app.

Which AI companion has the best memory?

Several apps make memory their main pitch. Nomi describes short, medium and long term memory, and Kindroid says its memory can recall hundreds and even thousands of earlier messages. No app remembers perfectly, and a claim on a home page is not a result. Run the 7-day test on two apps and trust your own scores over anyone's marketing, including ours.

Why does my AI companion forget things?

Most chat AI can only look at a limited amount of recent text at once, often called a context window. Older messages fall out of view unless the app saves the important parts separately as long-term memory. Forgetting usually means a fact never got saved, or it was saved but not brought back at the right moment.

Can I make an AI companion remember me better?

Often, yes. Be specific, repeat important facts in a natural way, and correct mistakes right away. If the app has a pinned memory, a memory list or character notes, put the facts that matter there instead of hoping the app picks them up. Short, clear statements save better than long rambling ones.

Does Aisha remember what I tell her?

Yes, remembering you is the point of her design. She keeps what you tell her and can follow up on it days later, like asking how an interview went. She will not be perfect, so test her the same way as any app: plant a small fact on day one and see what comes back. She works in your browser and on Telegram, and the free chat is limited before you pick a plan.

aisha
18+ only.© 2026 — all rights reserved