For yourself
A thinking tool only you use
A custom GPT loaded with your notes. Free with a ChatGPT subscription, private, and good enough for drafting and recall.
For customers
Answering repeat questions
A support chatbot trained on your docs. Built to deflect tickets, not to have a personality. Sounds like a company, because that is the job.
For your audience
A bot that represents you
A clone trained on how you explain things, living on your own site, answering as you when you are not there.
Try it freeWhat it actually is
The word chatbot covers two very different things, and picking the wrong one is the most common mistake.
The phrase people type, and slightly the wrong word. A chatbot answers on behalf of an organisation out of a shared knowledge base, and whatever personality it has is applied on top and could be swapped without changing what it knows. What you are describing is a persona: trained on one person's knowledge, frameworks and way of explaining, where the personality is not decoration but the thing being reproduced.
The distinction is worth holding on to, because it decides which tools can do the job. Anything built for deflecting support tickets is optimised to sound like a company. That is the right goal for a helpdesk and the wrong one for anything carrying your name. The longer version of this argument is in AI persona vs chatbot.
Whatever you call it, nobody is building you a bespoke language model. Every practical route in 2026 uses an existing model and points it at your content, a technique called retrieval. When someone asks a question, the system finds the relevant passage from your material and hands it to the model along with the question.
The consequence is liberating: you do not need a data science team, and you do not need thousands of examples. You need good source material and a clear sense of what the bot is allowed to say. For a longer treatment of how the retrieval side works, see how to train an AI clone.
What you need first
Gather this before you open any tool. It is the whole job.
Your explanations. Articles, course lessons, newsletters, internal docs. Anywhere you have written down how something works.
Your talks. Podcast episodes, webinars, recorded calls, YouTube. Transcripts of you speaking carry tone that writing often sands off.
Your repeat answers. The twenty questions you answer every week. This is the highest value material you own and it is usually sitting in your sent folder.
Your positions. The things you disagree with. A bot that cannot take your side of an argument does not read as you.
If you have a website with substantial writing on it, you already have enough to see whether this works. Personify builds a working preview from a URL in about 30 seconds, with no signup, which is the cheapest way to find out before you invest an afternoon.
The three routes
Same underlying technology. Very different results, depending on who the bot is for.
| Custom GPT | Support chatbot | Clone platform | |
|---|---|---|---|
| Best for | Private use | Ticket deflection | Representing you |
| Lives | Inside ChatGPT | Your help widget | Your site and links |
| Audience needs an account | Yes | No | No |
| Tone work | Prompt only | Minimal by design | Core of the product |
| You see the questions | No | Yes | Yes |
| Typical cost | $20/mo ChatGPT Plus | Varies by volume | Free tier, Pro from $29/mo |
The middle column is where most people get stuck. Support chatbot builders are excellent at what they were designed for, which is answering "where is my order" without a human. Point one at your body of work and you get accurate answers in a voice that belongs to nobody. That is a feature for a helpdesk and a problem for a personal brand.
How to build one
Assuming you want the third route, the one where the bot answers as you.
Gather, do not write
Resist the urge to write fresh content for the bot. You will write marketing copy, and it will sound like marketing copy. Pull what already exists: the article where you explained your framework properly, the podcast where you got asked a hard question, the long email you sent a client at 11pm.
Load it and let retrieval do the work
Upload the material rather than pasting summaries into an instruction box. A prompt has a size limit and no memory of what it left out. Retrieval pulls the right passage at the moment a question arrives, which means your bot can hold far more than any prompt could carry.
Interrogate it with real questions
Take your twenty most common questions and ask every one. Read the answers as if you were the person who asked. Wrong answers are almost always a hole in the source material rather than a fault in the model, so fix them by adding the missing explanation, not by adding adjectives to the prompt.
Decide what it must not do
Write down what the bot should refuse and what it should escalate. Anything where being confidently wrong is expensive belongs on that list. Give it a clean handoff so a person asking something serious reaches you rather than getting a guess.
Put it where people already look for you
A bot nobody can find gets no use. The obvious places are your site, the link in your bio, your course platform, and the reply you currently type by hand for the tenth time. Personify supports 100+ languages, which matters more than people expect once a bot is public.
Making it sound like you
The part everyone underestimates, and the only part your audience notices.
Accuracy is table stakes. Any of these tools will get the facts roughly right if you feed them decent material. What separates a bot people keep talking to from one they abandon after two messages is whether it answers the way you would, which is a much narrower target.
Voice lives in specifics: the analogy you always reach for, the caveat you always add, the question you ask back before answering, the thing you refuse to recommend. None of that survives being summarised. It has to come from examples of you doing it.
Tone comes from examples, not from adjectives. Writing "friendly and professional" in a prompt does almost nothing. Twenty of your real answers does almost everything.
When it works, people notice. Career coach Lucy Gilmour built a clone after 2,000 DMs arrived in 48 hours:
Conversations handled in 9 days
Revenue in the first 24 hours
Available without her
"The moment she nailed my tone, I got chills. It didn't just feel like some AI bot, it felt like I was talking to myself."
Where people go wrong
Four failures that account for most abandoned bots.
Feeding it the brochure. Marketing copy is the least representative writing you own. It produces a bot that sells instead of explaining, which is exactly what people came to avoid.
Prompting instead of sourcing. Hours spent rewording instructions when the real fix was adding the article that answers the question.
No boundaries. A bot with no refusal list will confidently answer a medical question at 2am. Decide the edges before it is public, not after.
Launching and leaving. The transcript of what people actually asked is the most useful thing the bot produces. Reading it weekly is what turns a demo into something worth having.
FAQ
Can I make an AI chatbot of myself for free?+
Yes, up to a point. Personify has a free tier, and you can preview a clone built from your website in about 30 seconds without signing up. Free tools are enough to see whether the idea works for you. What free versions generally do not include is the volume, integrations, and refinement you need once real people are using it every day.
How much source material do I need?+
Less than people expect for a usable bot, more than they expect for a convincing one. A few long articles or a couple of hours of transcribed talks will produce something that works. The difference between a bot that answers and a bot that sounds like you is usually depth on the questions you get most often, not total word count.
Will it sound like me?+
It will sound like your source material. If you feed it polished marketing copy, it will sound like marketing copy. If you feed it the way you actually explain things to a client, including the analogies and the caveats you always add, it will sound much closer to you. Tone comes from examples, not from adjectives in a prompt.
Can it use my voice?+
Voice cloning is part of Personify's done-for-you builds, not the self-serve tiers. Self-serve clones on the free and Pro plans are text. If you want the bot to speak in your voice, that is scoped on a discovery call. See DIY or done-for-you for the full comparison.
Is a chatbot of yourself the same as a custom GPT?+
Not quite. A custom GPT is a configured version of ChatGPT that lives inside ChatGPT, which means your audience needs a ChatGPT account to use it and you see nothing about who asked what. A clone platform gives you a bot on your own site with your own analytics and integrations. For private use, a custom GPT is often enough. There is a fuller breakdown in how to make your own ChatGPT.
What should the chatbot refuse to answer?+
Anything where being wrong is expensive. Medical, legal, and financial specifics, anything about an individual's private circumstances, and anything you would normally want a conversation about before answering. Set those boundaries explicitly and give it a route to hand the person to you.
See what yours would sound like
Paste your website and Personify builds a working clone of you in about 30 seconds. No signup needed to preview it, and no commitment if it is not for you.
Free tier available. No credit card needed.