The short answer
- An AI personal assistant (ChatGPT, Gemini, Claude, Copilot and the apps built on them) is very good at preparing: logging an expense from one sentence, setting a reminder, drafting a message or a quote, summarising a letter.
- It can also be confidently wrong: in a 2025 study by 22 public broadcasters, 45% of AI assistants’ answers about the news had at least one significant issue.
- The rule that matters most: it prepares, you confirm. Nothing should be sent, paid or saved without your say-so.
- For health, legal and investment decisions, use it to organise your questions, not to decide.
- Before you trust one with your life admin, ask five privacy questions, and never type a password, PIN or full card number into it.
What an AI assistant actually is
Today’s assistants run on large language models. They don’t look facts up in a database; they generate, word by word, the most likely continuation of the text so far, based on what they learned in training. Most of the time that produces useful, fluent answers. Some of the time it produces fluent answers that are false, known as hallucinations.
Why don’t they just say “I don’t know”? In a September 2025 paper, researchers from OpenAI and Georgia Tech (Kalai and colleagues) compare the model to a student facing a hard exam question: it guesses rather than admit uncertainty, because training and benchmark scoring reward a confident answer over an honest blank. That’s not a one-off bug; it comes from how these systems are built and graded.
So keep two things apart. An assistant is excellent at working with text you give it: your notes, your numbers, your draft. It’s far less reliable at stating facts it doesn’t have in front of it.
How many people already use one
In the US, about half of adults (49%) now use AI chatbots, up from a third in 2024, according to a Pew Research Center survey of 5,119 adults in February 2026. Roughly a quarter use them daily. The most common uses were searching for information (42%) and, among workers, tasks at work (38%). One in five (20%) said they use chatbots to get medical advice.
People use them and distrust them at the same time. In the same survey, 71% of Americans said the increased use of AI will make their personal information less secure; just 3% said more secure.
What it does well, every day
The best uses share a pattern: you supply the facts, it does the formatting, you check at a glance.
- Logging spending in one sentence. “$18.50 on lunch, cash” becomes a line with an amount, a category and an account. You can see instantly whether it’s right. This matters most for cash, which is what people forget (see how to record cash purchases).
- Setting reminders. “Remind me to pay the council tax on the 1st of every month” is easy to check: date, repeat, done.
- Remembering things. “Remember that Kofi’s shoe size is 12” stores a fact you already know, so you can find it without digging.
- Drafting. A message to your landlord, a polite chaser for a late payment, a quote for a £640 kitchen job: the assistant gives you a clean first draft, you fix the tone and the numbers (see quotes and invoices from your phone).
- Summarising and rewording. A dense letter from your insurer becomes three clear points. Keep the original next to it to check dates and amounts.
- Translating a short message for a relative or a client.
What it does badly: stating facts
The errors aren’t rare, and they don’t announce themselves. They come in the same confident tone as the correct answers.
- News. In the News Integrity in AI Assistants study (EBU and BBC, October 2025), journalists from 22 public service media organisations in 18 countries evaluated 2,709 answers from the free versions of ChatGPT, Copilot, Perplexity and Gemini. 45% had at least one significant issue; 31% had sourcing problems; 20% had accuracy problems, such as invented or outdated details.
- Law. Researchers (Dahl and colleagues, 2024) asked models specific, verifiable questions about US federal court cases. The models gave wrong answers 58% of the time (GPT-4) to 88% of the time (Llama 2). In Mata v. Avianca (2023), a New York federal judge fined two lawyers and their firm $5,000 after they filed a brief citing court decisions ChatGPT had made up.
- Health. In a randomised trial published in Nature Medicine (Bean and colleagues, 2026), 1,298 UK adults worked through medical scenarios. Tested alone, the models named the relevant conditions in 94.9% of cases. But people using those same models identified relevant conditions in fewer than 34.5% of cases, no better than a control group using whatever they’d normally use. The breakdown happened in the conversation: people left out details, and the models’ correct suggestions got lost.
The practical answer is to weigh two things: how much is at stake, and how easily you can check the result.
The golden rule: it prepares, you confirm
An assistant that can act (send, post, save, pay) is more useful than one that only talks, and riskier. The OWASP Top 10 for LLM applications, a widely used security reference, lists “Excessive Agency” as one of its ten risks. Its advice: use human-in-the-loop control so that a person approves high-impact actions before they happen. Its own example: an assistant with access to your email should have you review a message before it’s sent.
That’s how Binome360 works: your assistant prepares the record or document, shows it to you, and nothing is saved until you confirm.
$18.50 on lunch, paid cash
Ready: expense of $18.50, category Food, account Cash, today. Save it?
If the category or amount is off, you fix it before saving. Nothing is recorded without your OK.
Try Binome360 for freeTo be clear about what Binome360 doesn’t do: it doesn’t connect to your bank, never moves money, gives no investment advice and doesn’t analyse your spending on its own to send you alerts. It keeps your accounts, reminders, memory, quotes and invoices, under your control.
Money, health, law: where it shouldn’t decide
- Money. Use AI to keep track of spending, draft a budget or explain a term (“what’s an APR?”). Not to pick investments. In a joint alert in January 2024, the SEC, FINRA and NASAA told investors not to rely solely on AI-generated information for investment decisions, noting it can be “faulty, or even completely made up”, and to be wary of any claim that AI can guarantee returns. The same alert warns about AI-cloned voices of relatives asking for money; agreeing a family code phrase is one defence it suggests.
- Health. AI can help you prepare for an appointment: list symptoms, when they started, what you want to ask. Diagnosis and treatment belong with a doctor, nurse or pharmacist. In an emergency, call emergency services, not a chatbot.
- Law. AI can reword a contract or a letter so you understand it. Before you act on it (dispute, sign, go to court), check official guidance such as GOV.UK or your state’s court self-help pages, and talk to a qualified adviser.
Privacy: five questions to ask before you start
A personal assistant sees your spending, names, addresses and sometimes health worries. Ask first.
- Are my conversations used to train the model, and can I opt out?
- Where is my data stored, and for how long?
- Can I export and delete my data easily?
- Can the assistant act on its own, or does it ask me to confirm each action?
- Who else can see my conversations (a shared space with a partner, family or team)?
- I never type passwords, PINs or full card numbers into it
A few reference points:
- Training. Most major assistants have a setting to stop your chats being used to improve their models. In October 2025, France’s data protection regulator (CNIL) documented where it sits in ChatGPT (“Improve the model for everyone”), Claude (“Help improve Claude”), Copilot, Gemini and others. Check it the day you install.
- Promises are enforceable. The US Federal Trade Commission warned AI companies in January 2024 that breaking privacy commitments, for example a promise not to use customer data to train models, can violate the law, and that quietly changing terms of service is risky too.
- Your rights in the UK. Under data protection law you can ask an organisation for a copy of your personal data (a subject access request), ask it to delete your data, or object to how it’s used. The ICO says organisations usually have one month to respond to an access request.
Frequently asked questions
What is the best AI personal assistant app?
There’s no single winner, and models change every few months. Judge on your own use: does it do what you ask, does it show you what it’s about to do before acting, and what does it do with your data? A focused app (budgeting, reminders, invoices) is often more reliable for a specific job than a general chatbot.
Is ChatGPT a personal assistant?
It can play the part for drafting, summarising and answering questions. A personal assistant app usually adds memory of your own information, reminders that actually notify you, and records you can review, such as expenses or documents.
Can an AI assistant replace a financial adviser or a doctor?
No. It can help you understand, organise and prepare questions. In the Nature Medicine trial, people helped by an AI made no better medical decisions than people who weren’t. For investments, US regulators say not to rely solely on AI-generated information; for your health, see a clinician.
Why does AI make things up?
Because it generates the most likely text, not verified text, and because training and testing have long rewarded a confident answer over “I don’t know”. Ask for sources and check them yourself.
In short
An AI personal assistant earns its place through what it prepares: an expense logged, a reminder set, a clean draft. It stays fallible on facts, and it should never decide your money, health or legal questions on its own. Choose one that asks before it acts and is clear about what happens to your data.
First action: open the settings of the assistant you already use and check whether your conversations are used for training.
Sources
- Pew Research Center, “Americans and AI 2026: Chatbots, Smart Devices and Views on Impact”, 17 June 2026 (survey of 5,119 U.S. adults, 17–23 February 2026): pewresearch.org.
- EBU and BBC, News Integrity in AI Assistants, October 2025: ebu.ch/research/open/report/news-integrity-in-ai-assistants.
- Adam Tauman Kalai, Ofir Nachum, Santosh S. Vempala and Edwin Zhang, “Why Language Models Hallucinate”, arXiv:2509.04664, September 2025: arxiv.org/abs/2509.04664.
- Matthew Dahl, Varun Magesh, Mirac Suzgun and Daniel E. Ho, “Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models”, Journal of Legal Analysis, 16(1), 2024: arxiv.org/abs/2401.01301.
- Mata v. Avianca, Inc., 678 F.Supp.3d 443 (S.D.N.Y., 22 June 2023).
- Andrew M. Bean et al., “Reliability of LLMs as medical assistants for the general public: a randomized preregistered study”, Nature Medicine, 32, 2026: nature.com/articles/s41591-025-04074-y.
- OWASP GenAI Security Project, “LLM06:2025 Excessive Agency”: genai.owasp.org/llmrisk/llm062025-excessive-agency.
- SEC Office of Investor Education and Advocacy, NASAA and FINRA, “Artificial Intelligence (AI) and Investment Fraud”, investor alert, 25 January 2024: finra.org/investors/insights/artificial-intelligence-and-investment-fraud.
- Federal Trade Commission, Office of Technology, “AI Companies: Uphold Your Privacy and Confidentiality Commitments”, 9 January 2024: ftc.gov.
- Information Commissioner’s Office, “Getting copies of your information (subject access request)” and “Your right to get your data deleted”: ico.org.uk/for-the-public.
- CNIL, “IA et vie privée : comment s’opposer à la réutilisation de ses données personnelles pour l’entraînement d’agents conversationnels ?”, October 2025: cnil.fr.
Also available in Français.