Using AI chatbots and assistants well: what they are good at, where they fail, and protecting your privacy

The short answer

• AI chatbots and assistants are good at first drafts, short summaries, plain-language explanations and brainstorming. Treat what they write as a draft, not a final answer.
• They can be wrong while sounding sure, and can list sources that do not exist. Check anything that matters, such as money, law, health or a deadline, against the official source.
• Never type in passwords, one-time codes, account or Social Security numbers, or other people's private details. Look at the tool's settings for chat history and for whether your chats are used to improve the system.
• Scammers use AI to copy voices and faces. The FBI suggests a secret family word or phrase, and hanging up to call back on a number you look up yourself, before you send money.
• California law already covers some of this. Since 2019, a bot may not hide that it is a bot to sell you something or sway your vote. Since January 1, 2026, "companion" chatbots must disclose that they are AI and protect minors.
• For health questions, use the checking routine in Using an AI chatbot for a health question? How to check the answer.

This is general information, not legal, technical or medical advice. AI products change quickly, so check the current settings and policies of any tool you use.

What these tools are, in plain words

"Generative AI" is software that produces new text, pictures, audio or video. In a December 2024 warning, the FBI described it as technology that takes what it has learned from examples and "synthesizes something entirely new based on that information."

You may meet it in several forms:

• Chatbots: you type or speak a question and get a written answer.
• Assistants built into other products: in a phone, search engine, email program, word processor or photo app.
• Image, voice and video generators: they make pictures, speech or clips from a description.
• Companion chatbots: apps designed to act as a friend, partner or confidant. California has a separate law for these (see below).

A chatbot writes by building a likely-sounding answer from patterns in what it learned. It does not check a fact the way a careful person would, and it cannot see your situation unless you describe it. That is why the habits in this guide matter.

Good uses, and how to ask

These tools save time on work where you can easily spot a mistake, or where a mistake costs little.

Five habits for better answers

1. Give context without private details. "I rent an apartment in California" helps; your address and account number do not.
2. Say what you want back: length, format, reading level or language.
3. Ask for sources and links to official pages, then open them yourself.
4. Ask what it is unsure about and what could change the answer.
5. Push back. If something looks wrong, say so and ask it to explain, but do not take a quick "You're right" as proof either.

Where they fail

Most problems fall into a few patterns. Knowing them makes errors easier to catch.

• Confident errors. A chatbot can state something false in the same smooth, certain tone it uses for something true. Fluency is not accuracy.
• Made-up sources. Answers can include quotes, studies, laws, case names or web links that do not exist, or that exist but do not say what the chatbot claims.
• Old or out-of-place information. Fees, deadlines and laws change every year, and an answer may describe another state or country. California rules often differ from federal ones.
• No view of your situation. It knows only what you type. A missing detail, such as your county, your income or the date on a letter, can change the right answer.
• Arithmetic and dates. Totals, percentages and day counts can be off. Recalculate anything you will act on.
• Agreeing too easily. A chatbot may go along with a wrong assumption built into your question.

For how these problems show up in health answers, and why a polished answer can still be wrong, see the companion guide on checking health answers.

Check before you rely on an answer

1. Ask what is at stake. Party ideas need no checking. Anything about money, a contract, taxes, immigration, a court case, your health or a deadline does.
2. Open every source it gives. If a link is dead, goes somewhere else, or does not contain the claim, treat the whole answer with suspicion.
3. Go to the primary source: the agency's own website (ending in .gov or ca.gov for California agencies), the law itself, your bill, your lease, your policy or plan documents.
4. Check the date and the place. Is the page current? Is the rule for California, and for your city or county?
5. Ask a responsible person when it matters: the agency, your landlord or lender in writing, a legal aid office, a tax preparer or your own clinician.

Privacy: what to keep out of the chat

Assume that what you type into a chatbot, or say to a voice assistant, may be stored by the company and read later. Write as if a stranger might see it.

Never paste or type:

• Passwords, PINs, one-time sign-in codes or security answers.
• Bank, card or loan account numbers; your Social Security, driver's license, passport or Medicare number.
• Other people's private information: their health, finances, immigration status, addresses, phone numbers or photos, especially of children.
• Whole documents with identifiers, such as tax returns, medical records, court papers or pay stubs. Copy only the part you need, with names and numbers removed.
• Confidential work information, unless your employer has approved that tool for that use.

Settings and policies to check, in the app's settings and privacy policy, before you use a tool regularly:

• Are chats saved? Can you delete them, one by one or all at once?
• Are your chats used to train or improve the system, and can you switch that off?
• Is there a temporary or private chat mode that is not saved to your history?
• Can company staff review conversations, for example for safety?
• Does the assistant "remember" details about you between chats, and can you see and erase what it remembers?
• Is it connected to your email, calendar, files, contacts or location? Turn off connections you do not need.

Look again after a big app update, since settings and policies can change. On a shared or family device, sign out or use a separate profile so others cannot read your chats. Health questions raise extra privacy issues, covered in the health-answers guide.

AI in scams: cloned voices and deepfakes

The FBI's December 3, 2024 alert says criminals use generative AI to make fraud more believable, and lists how:

In May 2025 the FBI also warned of text and AI-generated voice messages pretending to come from senior U.S. officials. Its advice: "do not assume it is authentic."

If you get a panicked call or message

1. Ask for your family's secret word or phrase. The FBI suggests agreeing on one ahead of time. Set it up today, in person, and do not post it anywhere.
2. Hang up and call back on a number you already have or look up yourself, not one the caller gives you.
3. Listen closely to the tone and word choice. Does it sound like how this person really talks?
4. Do not send money, gift cards or cryptocurrency to anyone you know only online or by phone, and do not share sensitive information with them.
5. If you already paid, call your bank or payment company at once and ask it to stop or recall the payment, then report it (see the last section).

Make yourself harder to copy. The FBI suggests limiting the photos and recordings of your voice that are public, making social media accounts private, and accepting only followers you know. For text-message scams and phone scams in general, see Fake delivery, toll and bank texts and Verify before you pay.

Sexual deepfakes. A 2025 California law, AB 621, lets victims, including minors, seek civil relief of up to $250,000 per action against third parties who knowingly help distribute nonconsensual sexually explicit material, according to the Governor's office.

Checking AI images and videos

The FBI says "it can be difficult to identify when content is AI-generated." Its tips for spotting fakes are to look for subtle flaws:

• Distorted hands or feet; unrealistic teeth or eyes; blurry or irregular faces.
• Glasses or jewelry that look wrong; shadows that fall the wrong way; watermarks.
• In video: lag, a voice that does not match the lips, and unnatural movement.

A convincing fake may show none of these, so also check the story:

1. Who posted it first? Look for the original account. Is it the real organization or person, or a look-alike?
2. Is anyone else reporting it? Real news from a disaster, an arrest or a celebrity usually appears in several established news outlets and on official accounts.
3. Search for the picture itself. A reverse image search can show where else it has appeared and whether it is older than claimed.
4. Look for a label, such as "AI-generated" or "made with AI." But do not treat a missing label as proof that something is real.
5. Slow down when an image or clip makes you angry or afraid, or asks for money or a share. That is what fakes are built to do.

California's labeling rules. California's AI Transparency Act sits in Business and Professions Code sections 22757 to 22757.6. It was amended in 2025 (AB 853) and again in 2026 (AB 2713 and SB 1000). The Governor's office says the 2026 changes are meant "to ensure Californians can identify AI-generated material," and describes a ban on deleting digital watermarks. Separately, a March 30, 2026 executive order (N-5-26) directs the state Department of Technology to recommend best practices for watermarking AI-generated images and manipulated video. Labels will not be on everything you see, so keep checking the story behind an image.

California laws that touch consumer AI

This table summarizes laws as described in the law's text or by the state agency or Governor's office named in Sources. Read the law itself for exact duties. Under the California Constitution (article IV, section 8), an ordinary statute from a regular session takes effect on the January 1 after a 90-day period from enactment, so most laws signed in September 2026 start January 1, 2027, unless the law sets a later date. Bills passed in 2025 and signed that fall generally took effect January 1, 2026.

What the bot law means for you. Section 17940 defines a bot as "an automated online account where all or substantially all of the actions or posts of that account are not the result of a person." As written, section 17941 targets bots used to mislead people about their artificial identity to push a sale or sway a vote, and says the disclosure must be "clear, conspicuous, and reasonably designed" to tell people it is a bot. It does not reach every automated account in every situation. If an account selling you something insists it is a person and you suspect otherwise, stop and buy through the company's own website or store instead.

A chatbot that claims to be a therapist or doctor. The California Board of Psychology says AB 489 treats AI that presents itself as holding a Board license, or uses reserved titles, as unlicensed practice, and that the Board can respond with cease-and-desist letters, citations or discipline. You can report such a tool to the licensing board for that profession.

Children, teens and AI companions

Companion chatbots are built to feel like a friend. That can be comforting, and it is also why California now regulates them. Under SB 243, in effect since January 1, 2026, a companion chatbot platform must tell users that the conversation is artificially generated. For minors, it must give break reminders and keep the chatbot from showing sexually explicit images. All platforms must have protocols for users who express suicidal thoughts or self-harm, and must report how often they sent users crisis-service notifications, as the Governor's office describes it.

On September 10, 2026, the Governor signed 13 child online-safety bills. They include SB 1119 on companion chatbots and children's safety, which the Governor's office calls "Adam's Law." It adds crisis protocols for suicidal thoughts, parental controls, notification "should a child disable safety settings," independent child-safety audits and annual risk assessments, and SB 867 on toys with companion chatbots. The release gives no start dates; under the usual rule, most will begin January 1, 2027.

What parents and caregivers can do now:

• Ask which AI chatbots your child uses, including ones built into games, social media and smart toys.
• Check the app's age rating and set up parental controls on the device and in the app. See Before you give a child a new app.
• Explain, in plain words, that a chatbot is not a person, does not love them, can be wrong, and keeps what they type.
• Agree that they never share their full name, school, address, photos or passwords with a chatbot, and that they tell you if one asks.
• Notice changes: hours of secret chatting, pulling away from friends and family, distress when cut off, or talk of self-harm. Talk with your child's clinician about worries like these.

If a child or teen talks about suicide or self-harm, call 988, the Suicide and Crisis Lifeline, or 911 if they are in immediate danger.

Where to get free help

• Report fraud, including AI voice or video scams, to the FBI's Internet Crime Complaint Center at ic3.gov. If you are 60 or older and need help filing, the FBI Elder Fraud Hotline is (833) FRAUD-11, or (833) 372-8311, Monday to Friday, 10 a.m. to 6 p.m. Eastern.
• If money left your account, call your bank first. The FBI says to ask for a recall of any wire transfer as quickly as possible.
• If your identity was stolen, the FBI also points you to IdentityTheft.gov.
• An AI tool posing as a licensed professional: the California licensing board for that profession, such as the Board of Psychology.
• Local help with bills, food or housing: dial 211.
• Crisis: 988 for the Suicide and Crisis Lifeline, which the Federal Communications Commission designated as its three-digit number; 911 for any emergency.
• Related guides: checking AI health answers, fake delivery, toll and bank texts, verify before you pay, and, for the policy detail on 2026 AI laws, California's September 30 AI package.

Sources

1. Criminals Use Generative Artificial Intelligence to Facilitate Financial Fraud (Alert I-120324-PSA) — Federal Bureau of Investigation, Internet Crime Complaint Center, December 3, 2024.
2. Senior US Officials Impersonated in Malicious Messaging Campaign (Alert I-051525-PSA) — Federal Bureau of Investigation, Internet Crime Complaint Center, May 15, 2025.
3. Elder Fraud — Federal Bureau of Investigation, Internet Crime Complaint Center, accessed October 29, 2026.
4. California Business and Professions Code § 17940 (2025 edition) — Justia, reproducing the California Code, accessed October 29, 2026.
5. California Business and Professions Code § 17941 (2023 edition) — Justia, reproducing the California Code, accessed October 29, 2026.
6. California Business and Professions Code, Division 8: Special Business Regulations (chapter list) — Justia, 2025 California Code, accessed October 29, 2026.
7. California Constitution, Article IV, Section 8 — Justia, accessed October 29, 2026.
8. Governor Newsom signs bills to further strengthen California’s leadership in protecting children online — Office of the Governor of California, October 13, 2025.
9. AB 489 (Bonta) Chapter 615, Statutes of 2025 – Health care professions: deceptive terms or letters: artificial intelligence — California Board of Psychology, operative date January 1, 2026.
10. Governor Newsom signs SB 53, advancing California’s world-leading artificial intelligence industry — Office of the Governor of California, September 29, 2025.
11. As Trump rolls back protections, Governor Newsom signs first-of-its-kind executive order to strengthen AI protections and responsible use — Office of the Governor of California, March 30, 2026.
12. Governor Newsom signs first-in-the-nation AI safeguards to protect Californians, calls on the federal government to do its part — Office of the Governor of California, September 9, 2026.
13. Governor Newsom signs the strongest child safety chatbot and social media laws in the nation — Office of the Governor of California, September 10, 2026.
14. California’s nation-leading AI framework just got stronger, Governor Newsom signs more first-in-the-nation worker protections and more — Office of the Governor of California, September 30, 2026.
15. Abbreviated Codes — North American Numbering Plan Administrator, accessed October 29, 2026.