iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
The claim that AI ignores what you told it 90% of the time is not backed by any published study, so this article does not treat it as fact. What the benchmarks do support is narrower: assistants can lose details across long, multi-session conversations, can recall a fact without using it during a task, and can carry stored memories into places they do not belong. A single memory shared across assistants may reduce some of that friction, but none of the sources cited here tests whether it fixes most of it. Treat it as a product idea until it has been measured.
Why does my AI keep forgetting what I told it?
“Memory” is not one skill for an assistant. LongMemEval, a benchmark published at ICLR 2025, splits it into five abilities:
“We introduce LongMemEval, a comprehensive benchmark designed to evaluate five core long-term memory abilities of chat assistants: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention.”
PerformancePC Slower Than It Used to Be?DriversCrashes, No Sound, or Screen Glitches?PerformanceWindows Errors? Fix Them Before They SpreadSpecial offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.
- Information extraction: pulling out a detail you mentioned earlier.
- Multi-session reasoning: combining facts that came up in separate conversations.
- Temporal reasoning: working out when something happened or the order of events.
- Knowledge updates: replacing an old fact with a newer one, such as a new city, rather than keeping both as current.
- Abstention: saying it does not know rather than inventing an answer.
A product can be strong at one of these and weak at another, so “it forgot” can describe several different failures. Knowing which one you are seeing tells you what to fix.
#1 Best Overall
- ENHANCED CONTEXT WITH MULTIMODAL INPUT: Capture audio, type notes, add images, and press to highlight key moments for richer context. During recording, instantly mark key moments with a single button press. Simultaneously enrich your audio by snapping photos of important documents or typing in ideas
- CHAT WITH YOUR RECORDINGS USING "ASK Plaud": Unlock deeper insights with this interactive AI. Ask questions, extract key points, draft emails, and get next-step suggestions—all grounded in your original audio for reliable, ready-to-use answers
- INTELLIGENT RECORDING WITH AI DIRECTIONAL AUDIO: Enjoy seamless, intelligent recording with Plaud Note Pro. Its AI automatically switches between call and meeting modes while recording, while directional audio and real-time spatial awareness minimize noise to capture voices with crystal clarity
- Everything Included: Includes Plaud Note Pro, magnetic case, magnetic ring, charging cable, and a free Starter Plan with 300 transcription minutes per month. Upgrade anytime in the Plaud app to Pro Plan (1,200 min/mo) or Unlimited Plan(Up to 24 hours of transcription per user per day)
- PREMIUM ULTRA-SLIM DESIGN WITH INSTANTVIEW DISPLAY: Meticulously designed, the AI Note Taker is just 0.12 inches thin and 1.06 oz —about the size of a credit card. Its sleek aluminum body with a textured wave finish features a vivid AMOLED display, letting you check battery and recording status at a glance, while it seamlessly works with Apple Find My to ensure you never misplace it
What the benchmark numbers do and do not show
Each benchmark measures a different slice of memory, and its headline figure applies only to its own test conditions.
| Benchmark (venue, date) | What it measures | Reported figure | What it does not show |
|---|---|---|---|
| LongMemEval (ICLR proceedings, 2025) | Five long-term memory abilities of chat assistants | A 30% accuracy drop on memorizing information across sustained interactions, reported for commercial chat assistants and long-context LLMs | A general rate at which assistants ignore instructions in everyday use |
| Mem2ActBench (ACL Anthology, 2026) | Active use of memory in tool-oriented tasks, such as filling in tool parameters | 2,029 synthesized sessions and 400 tool-use tasks; human evaluators judged 91.3% of the generated tasks strongly memory-dependent | Everyday chat behavior; the tasks are generated, not drawn from real use |
| PersistBench (PMLR proceedings, 2026) | Two memory-specific risks: cross-domain leakage and memory-induced sycophancy | 18 evaluated models; median failure rates of 53% on cross-domain samples and 97% on memory-induced sycophancy samples | How often these failures happen in everyday assistant use |
| TriggerBench (Microsoft Research, June 2026) | Prospective memory: spontaneously recalling and acting on constraints the user set earlier | Not stated as a numeric figure in the abstract; reported as harder than retrospective memory in its evaluation | A failure rate for any assistant |
| RHELM (Microsoft Research) | Realistic, heterogeneous, evolving long-horizon assistant memory | Not stated; the benchmark page describes its scope | Any results |
Remembering a fact is not the same as using it
Asking an assistant “what is my budget?” tests recall. Using that budget without being told to, while choosing a tool or filling in a form, tests something else. The Mem2ActBench paper focuses on the second:
“Experiments on seven memory frameworks show that current systems remain inadequate at actively utilizing memory for parameter grounding, highlighting the need for more effective approaches to evaluate and improve memory application in task execution.”
Free tools Windows power users keep installed
One-click scans. No signup required.
Special offer. See more information about Outbyte and uninstall instructions. Please review EULA and Privacy policy.Rank #2
Plaud Note Pro AI Voice Recorder Transcribe & Summarize for Meetings Calls
- AI-POWERED TRANSCRIPTION & SUMMARIES: Plaud Note Pro is your professional voice transcriber, delivering high-accuracy transcription in 112 languages with auto speaker labels. Powered by top AI models and thousands of templates, Note Pro instantly creates structured summaries, mind maps, To-Do lists, and proposals tailored to your role and industry
- ENHANCED CONTEXT WITH MULTIMODAL INPUT: Capture audio, type notes, add images, and press to highlight key moments for richer context. During recording, instantly mark key moments with a single button press. Simultaneously enrich your audio by snapping photos of important documents or typing in ideas
- CHAT WITH YOUR RECORDINGS USING "ASK Plaud": Unlock deeper insights with this interactive AI. Ask questions, extract key points, draft emails, and get next-step suggestions—all grounded in your original audio for reliable, ready-to-use answers
- INTELLIGENT RECORDING WITH AI DIRECTIONAL AUDIO: Enjoy seamless, intelligent recording with Plaud Note Pro. Its AI automatically switches between call and meeting modes while recording, while directional audio and real-time spatial awareness minimize noise to capture voices with crystal clarity
- Everything Included: Includes Plaud Note Pro, magnetic case, magnetic ring, charging cable, and a free Starter Plan with 300 transcription minutes per month. Upgrade anytime in the Plaud app to Pro Plan (1,200 min/mo) or Unlimited Plan(Up to 24 hours of transcription per user per day)
The benchmark is built mostly from cases where memory is needed, since human evaluators judged 91.3% of its generated tasks strongly memory-dependent.
Recall on request
This is what most people test first, and it is the easiest to check. It corresponds to the information-extraction ability in LongMemEval. Passing it says little about the next two cases.
Applying a stored detail during a task
Parameter grounding means filling a tool’s required fields with the correct stored details, such as a saved shipping address or a default account. An assistant can answer “what is my address?” correctly and still enter the wrong one when it books something. That gap is the one Mem2ActBench targets.
Rank #3
- YOUR AI PERSONAL ASSISTANT FOR EVERYDAY PRODUCTIVITY: More than a voice recorder, Pocket works as your AI personal assistant to capture, transcribe, and summarize meetings, calls, and ideas instantly. Core features are included out of the box, with optional advanced tools available for power users.
- ONE-TAP RECORDING FOR REAL-LIFE MOMENTS: Capture meetings, phone calls, and in-person conversations instantly with a simple tap, no typing, no interruptions, just effortless note-taking anywhere you go.
- SMART AI INSIGHTS & ORGANIZATION: Pocket automatically turns recordings into clear summaries, key action items and structured conversation maps so you can quickly review what matters without digging through audio.
- TURN CONVERSATIONS INTO ACTION WITH “ASK POCKET”: Don’t just record, understand. Instantly ask questions across your meetings, extract key insights and generate next steps in seconds. All grounded in your recordings, so answers stay accurate and reliable.
- MAGSAFE COMPATIBLE FOR SEAMLESS USE: Easily attach Pocket to your iPhone or other MagSafe compatible devices for convenient, hands-free recording on the go. Perfect for capturing meetings, calls, and ideas without needing to hold your device.
Bringing it up unprompted
TriggerBench, published by Microsoft Research in June 2026, studies prospective memory: spontaneously recalling and acting on a constraint you set earlier, without being asked. Its abstract reports that this is harder than retrospective memory in its evaluation. The everyday version is an assistant that drafts a formal email after you said you wanted casual drafts, unless you remind it first.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →How do I get AI to remember my preferences?
No method works the same way across every product, and none of the benchmarks above ranks consumer assistants. What you control is how you store preferences and how you check them.
- Use the assistant’s memory or custom-instructions settings for stable preferences such as tone, output format, and units. Menu names and locations differ by product, so look for the memory or personalization section in the app you use.
- Write preferences as checkable rules. “Use metric units and bullet lists” can be verified in a reply; “be concise” cannot.
- For tasks that matter, restate the one or two constraints that cannot be missed at the start of the request.
- Keep a short preferences note you can paste into a new conversation when stored memory is unavailable or unreliable.
A simple test you can run on any assistant
- Give the assistant three preferences you can check, such as a required output format, a units rule, and one fact about your work.
- Confirm they are stored, using whatever memory view the product offers.
- Open a new conversation without mentioning the preferences. Ask for a task where all three should appear, such as drafting a short status report.
- Score each preference three ways: recalled when asked, applied without a reminder, and not wrongly applied to an unrelated task.
- Repeat after several sessions and note any preference that dropped out or was applied to the wrong task.
Why more memory is not automatically better
Persistent memory has costs as well as benefits. PersistBench names two memory-specific risks:
Rank #4
- Plaud Intelligence: Capture conversations in 112 languages and generate accurate transcripts with the Plaud App and Web. Plaud Intelligence uses leading models like GPT-5.5, Claude Sonnet 4.6, and Gemini 3.1 Pro to transform raw audio into structured insights. Choose from over 10,000 professional templates to generate mind maps and to-do lists, turning hours of discussion into immediate clarity
- Multiple Ways To Wear With Included Accessories: Adapt Plaud NotePin S to any workflow instantly with four included accessories. Wear your device effortlessly as a necklace, wristband, clip, or pin. Plaud NotePin S features a dedicated physical record button for precise, tactile control. Stay professional and keep your intelligence within reach all day
- Enterprise-grade Privacy: Built to the highest standards with ISO 27001/27701, SOC 2, HIPAA, GDPR, and EN18031 compliance. Every conversation is secure and protected. It is the trusted choice for creative, medical, and business professionals handling sensitive info
- Multimodal Input & Multidimensional Summaries: Capture audio, type notes, add images, and press/tap to highlight for richer context with multimodal input. Press the record button to mark key moments in real time. Plaud transforms a single conversation into multiple perspectives, providing faster, clearer insights, and unifies these inputs to deliver role-specific summaries that reflect your intent and priorities
- Lightweight Power and Peace of Mind: Weighing only 0.61 oz, Plaud NotePin S delivers 20 hours of continuous recording and 40 days of standby time. Store up to 64GB of audio locally, ensuring you capture every insight even without an internet connection
“We identify two long-term memory-specific risks: cross-domain leakage, where LLMs inappropriately inject context from the long-term memories; and memory-induced sycophancy, where stored long-term memories insidiously reinforce user biases.”
On the benchmark’s samples, median failure rates across 18 models were 53% for cross-domain leakage and 97% for memory-induced sycophancy. In everyday terms, a memory that carries a work project into a personal conversation, or that keeps agreeing with a view you once expressed, makes an assistant less reliable even when it seems more personal.
Forgetting matters too. A stored fact that is no longer true, such as an old job or an outdated address, can be applied as if it were current. This is why the knowledge-update ability in LongMemEval is tested separately from simple recall.
Can one memory work across different AI assistants?
The sources do not establish that it can, or that it would fix most instruction failures. A shared memory layer is a sensible design idea, because it would separate what you have told an assistant about yourself from any single app. But it inherits every problem above: it has to recall, apply, update, and forget correctly, and it must avoid leaking context across domains. RHELM, Microsoft Research’s benchmark for realistic, heterogeneous, evolving long-horizon assistant memory, describes the kind of testing a cross-assistant claim would need. None of the sources cited here tests a single memory shared across named assistants.
If you are evaluating a product that claims this, check the following:
Quick Recap
- What it stores: stable preferences, one-off details, or inferred traits, and whether you can see which is which.
- Control: whether you can inspect, edit, delete, and correct stored details.
- Updating and forgetting: whether old or uncertain information can be withheld or removed, and whether a newer fact replaces an older one.
- Assistant coverage: which assistants and models it works with, confirmed in the vendor’s current documentation, since feature sets change.
- Application: whether it applies preferences correctly during tasks rather than only retrieving them, tested with the five-step check above.
- Data handling: where your data is stored and how it is used. Read the vendor’s policy directly; this article does not assess it.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
