How the AI works

Can AI Actually Remember Past Conversations With You?

> ANSWER[AEO]

Whether an AI genuinely remembers something you told it weeks ago — not just within the current chat — used to be an open, mostly untested question. It now has a name, a standard benchmark, and a growing body of published comparisons: it is measurable, and current best approaches clearly outperform naive alternatives, though no system gets it perfectly right yet.

How do you even test whether an AI remembers?

The field's standard evaluation is LoCoMo — a benchmark of roughly 1,540 questions spanning four categories, purpose-built to test recall across multiple separate conversation sessions rather than within one continuous chat.

That distinction is the entire point of the benchmark: a system with a merely long context window can look like it remembers as long as everything fits in one conversation, but LoCoMo specifically tests whether a fact survives being retrieved from an earlier, distinct session — closer to how an actual ongoing relationship with a person works.

What happens if you just feed the whole conversation history back in every time?

This is the naive baseline every memory architecture gets compared against, and the research is fairly blunt about its limits. One head-to-head comparison of ten different approaches — spanning published research baselines, open-source tools, retrieval-augmented generation, full raw-context reloading, and dedicated memory systems — found full-context approaches averaging around 26,000 tokens per conversation just to answer a query, versus roughly 7,000 tokens for a more selective, retrieval-based memory approach on the same benchmark.

That's not a marginal difference — it's the gap between an approach that's economically sustainable at scale and one that isn't.

Do these systems actually work reliably, or just better than nothing?

The honest answer is "meaningfully better than nothing, but not solved." Published comparisons on LoCoMo show retrieval-based and graph-based memory architectures consistently ahead of full-context reloading on both accuracy and cost — but the same body of work also documents real failure modes: privacy risks in what gets stored, memory contamination where incorrect facts get incorporated and persist, and inconsistent performance on harder multi-hop questions that require combining several remembered facts rather than recalling one in isolation.

That combination — clearly better than the naive alternative, clearly not a finished problem — is a more accurate picture than either "AI has real memory now" or "AI can't remember anything," and it's worth holding onto that nuance rather than overclaiming in either direction.

What should you actually expect from an AI tool that claims to have memory?

Based on where the published research currently stands:

For the architecture behind all of this, see how AI memory actually works; for why the "just use a longer context window" alternative degrades, see context rot.

Key takeaways

  • > LoCoMo, roughly 1,540 questions, is the benchmark that tests recall across separate sessions.
  • > Reloading the full history costs around 26,000 tokens per conversation versus roughly 7,000 for selective retrieval.
  • > Retrieval and graph-based memory beat full-context reloading on both accuracy and cost.
  • > Known failure modes remain: contaminated facts, and weak multi-hop recall.
  • > Real memory updates and resolves contradictions; it does not just accumulate.

Sources

Frequently asked questions

How do researchers measure whether an AI "remembers" something?
The standard benchmark is called LoCoMo, a set of roughly 1,540 questions built specifically to test recall across multiple separate conversation sessions, not just within a single long chat.
  • A long context window alone won't pass it. The information being tested has to survive being retrieved from an earlier, separate session entirely.
  • It is the field's way of comparing memory architectures head to head — see how AI memory actually works for what those architectures do.
Which memory approach performs best on these benchmarks?
Published comparisons vary by method and continue to shift as new architectures are released, but hierarchical, graph-based, and selective-retrieval approaches have generally outperformed simply dumping an entire raw conversation history back into context.
  • The advantage shows up twice: in accuracy and in token cost.
  • The baseline they beat is full raw-context reloading, which also degrades in reliability as input grows — see context rot.
Is this the same as an AI having a personality it remembers?
Not quite — that's closer to persona consistency, a related but separate design goal.
  • The memory research discussed here is about factual recall: correctly retrieving something specific a person said, or a fact that was established.
  • The time span is the point — potentially weeks or months earlier.