What Does an AI Assistant Remember About Your Customers?
The short answer
It depends on the plan. Every plan follows the conversation currently open. Paid plans also carry what that same person said to you in earlier conversations in the same app, so a returning customer does not have to repeat themselves. What it never does is guess at things you never told it.
The WhatsAssist team ·
Memory is the difference between a reply that sounds like your business and one that sounds like a form. A customer who told you their size last month does not want to be asked again, and being asked again is the moment they remember they are talking to software.
So it is worth knowing exactly what is carried and what is not, because the honest answer has limits in it, and the limits are the part most tools will not tell you.
What is remembered inside one conversation?
Everything said in it, on every plan including Free.
That is the baseline, and it matters more than it sounds. Most customer exchanges are a handful of messages over ten minutes: a question, a clarification, a decision. A reply that has read the whole of that thread already avoids the worst failure, which is answering message four as though messages one to three never happened.
If you never pay for anything, this is what you get, and for a business whose customers mostly buy once it may be all you need.
What is remembered between conversations?
On paid plans, what that same person said to you before.
This is the part people are actually asking about. Somebody messaged you in June, bought a jacket, mentioned they were buying for their brother. They come back in September. On Free, that is a blank thread. On a paid plan, the reply can draw on the earlier conversation, so they do not start from nothing.
Where this earns its money is not novelty, it is repetition. A regular customer asking their usual question, a buyer following up on something they ordered, a complaint that references a previous complaint: those are the moments where starting from zero is visibly worse.
The top plans add one more thing, which is your writing voice, learned from messages you actually sent rather than from a tone setting you picked off a list. The plans and what each carries are on the pricing page.
What does it deliberately not do?
This is the section worth reading twice, because these are real limits rather than features we have not got to yet.
| What people assume | What actually happens |
|---|---|
| It links the same person across different apps | It does not. Somebody on Instagram and the same person on WhatsApp are two separate histories |
| You can remove one detail it has picked up | There is no per-item control. Deleting your account deletes everything, and that is the only removal there is |
| Each reply shows you what it drew on | No reply carries any indication of where its context came from |
| Customers get labelled automatically by how they behave | Nothing is assigned from behaviour. Groups match on things you set yourself |
| It fills gaps with sensible guesses | It does not invent a price, a stock level or a policy you never gave it |
The first row has a whole article behind it, because it surprises people the most: the same customer on WhatsApp and Instagram explains why joining them automatically would be worse than leaving them apart.
The last row is the one that makes the rest safe. Memory that fills in blanks confidently is more dangerous than no memory at all, because a wrong recollection sent in your name is a promise your customer will hold you to.
How long is any of it kept?
Until you delete it. There is no schedule that quietly expires it, and if that ever changes it will be said on the privacy page before it happens rather than after.
Two things it is never used for: it is not used to train a model, and it is not sold. Those are worth stating plainly because they are the two questions most owners have and most tools answer vaguely.
Should you pay for memory between conversations?
Ask one question about your own inbox: do the same people come back?
If most of your customers buy once and never write again, the between conversations part earns you very little, and the free tier’s thread memory is doing the real work. A market stall, a one-off repair, a tourist-facing business.
If people return, it is the single upgrade that changes how the replies read. A tailor, a clinic, a supplier, a salon, anywhere the second conversation with a person is a normal event rather than a surprise. There the cost of starting from nothing is paid on every repeat customer, every time.
There is a middle case worth naming. If your repeat business is real but small, you can get most of the benefit for free by keeping your own record, because a customer list you control does not depend on any tool and survives you changing one. What the assistant adds is that it applies at reply time without you looking anything up.
Where does memory matter most?
On complaints, and it is not close.
A customer complaining for the second time about the same thing is the worst possible moment to have forgotten the first. The reply that starts “sorry to hear that, can you tell me what happened” to somebody who told you exactly what happened three weeks ago is the reply that turns an annoyed customer into a public one. Replying to a customer complaint covers what a good complaint answer looks like, and continuity is half of it.
The corollary is worth saying too. Because complaints are where memory matters most, they are also where the limits above matter most, and it is why anything involving compensation or fault belongs with a person rather than with an automatic reply, whatever it remembers. What the product does and does not do is on the features page.
Frequently asked questions
Does it remember a customer who messages from a new number?
No. To the app that is a different person, so it starts fresh. If you need the history to follow them, connect it in your own records rather than expecting the app to work it out.
Can I see what it knows about a particular customer?
Not as a per-reply explanation. No reply tells you which part of the history it drew on, so if you need certainty about what a customer was told, read the conversation itself.
Does memory make replies slower?
Not in any way a customer would notice. The practical difference is in what the reply says, not how long it takes to arrive.
If I turn off automatic sending, does it still remember?
Yes. The switch controls whether a reply goes out by itself or waits for you in the app. It does not change what the reply was written from.
What happens to the memory if I change plan?
Moving up gives later replies more to work from. Moving down narrows what a reply is written from, and it does not delete anything you have already accumulated. If you want the history gone rather than unused, that is a deletion, and deletion removes everything rather than a slice of it.
Does it work on an iPhone?
No. WhatsAssist depends on Android notification access and requires Android 10 or newer. There is no iPhone version and no announced release date.