{"course":{"slug":"ai-for-small-business","title":"AI for a Small Business","url":"https://addaly.com/learn/ai-for-small-business","level":"beginner","track":"work"},"lesson":{"slug":"free-and-what-it-costs","title":"Free tiers, and what they take instead of money","url":"https://addaly.com/learn/ai-for-small-business/free-and-what-it-costs","position":79,"of":88,"minutes":9,"answer":"A free tier is paid for in data rights, rate limits or a smaller model.","keyIdea":"A free tier is paid for in data rights, rate limits or a smaller model, and which of the three it is determines whether it is fine for your marketing copy and unacceptable for your client files."},"body":{"format":"markdown","text":"## Free is a price, not an absence of one\n\nEvery free AI tier is funded somehow, and for a small business the useful question is not whether it is free but what it takes instead.\n\nThere are only three answers, and they have very different consequences.\n\n## One: your inputs train the model\n\nConsumer free tiers frequently reserve the right to use what you type to improve the service. Paid business and API tiers usually do not, and say so in the terms.\n\nTwo things follow. This is a real distinction and it is usually the main thing you get for the money on a business plan. And it is a setting as often as a plan: several products let you turn training off on a free account, in a preferences screen nobody visits.\n\nWhat it means practically: a customer's medical detail, a client's unannounced product, a supplier's confidential pricing, a colleague's grievance. Those do not go into a tier whose terms permit training, and the mitigation from the data lesson applies: strip identifying detail before pasting, and the question becomes much less sharp.\n\nCheck the terms, note the date you checked, and put that note beside the tool in your list. Terms change, and your obligation to your customers does not change with them.\n\n## Two: rate limits and quiet downgrades\n\nThe second currency is capacity. Free tiers cap messages per hour, files per day, or length. Fine for occasional work, and it fails exactly when you are busy, which is when you were relying on it.\n\nLess obviously, free tiers are often routed to smaller models, and during heavy demand some products silently route everyone to a smaller one. You will not be told. The symptom is a day when the output is noticeably flatter and nothing you changed explains it, which is the argument, from the logging lesson, for recording which model produced what.\n\n## Three: the free tier is the advert\n\nThe third kind is a genuine sample designed to get you onto a paid plan. Nothing sinister, and it has one operational consequence: free tiers get withdrawn or reduced with little notice, so anything your business depends on daily should not sit on one.\n\n## The free path\n\nDistinct from free tiers, and worth knowing because it is the only version that nobody can withdraw.\n\n**Open-weight models run locally.** Ollama or llama.cpp on your own machine, with a model like Llama, Mistral, Qwen or Gemma. A recent laptop with 16GB of memory runs a small model at usable speed. No per-use cost, no rate limit, and (this is the real argument) nothing leaves the building, which makes the confidentiality question disappear rather than being managed.\n\nThe honest limitations: a small local model is meaningfully weaker than a frontier model at reasoning and long documents, setup is an afternoon, and the electricity is not zero. Module one covered this door; here it is the answer to a specific question, which is what to do when the data is the constraint.\n\n**Free tools that are not AI at all.** LibreOffice, GIMP, Inkscape, Audacity, DaVinci Resolve's free edition, Python. Much of what gets bought as an AI subscription is a task these already do.\n\n**Open-weight models on somebody else's hardware.** A middle option that gets overlooked: several providers host open models at a fraction of frontier prices, charged per token. You get the low cost and the ability to move between providers, because the model is the same wherever it runs, which is the lock-in argument from later in this module, arriving early.\n\n## Reading the terms in five minutes\n\nYou are not reading the whole document. You are looking for four things, and a search box finds all of them.\n\nSearch for **train**. This tells you whether inputs may be used to improve the service, and whether there is an opt-out.\n\nSearch for **retain** or **delete**. How long do they keep what you send, and can you make them remove it.\n\nSearch for **sub-processor**. Who else touches the data, which matters for the contracts lesson later in this module.\n\nSearch for **region** or **transfer**. Where the processing happens, which matters in several regimes.\n\nWrite the four answers and the date on one line in your tools list. It takes five minutes per tool, once, and it converts a vague anxiety into a fact you can act on.\n\n## Choosing, in one line\n\nConfidential client material: paid business tier with training off, or a local model. Ordinary business writing and analysis: a free tier is fine. Anything the business depends on daily: pay, because dependence on something nobody owes you is a risk with no upside."},"check":{"question":"An architect uses a free chat tier to summarise planning documents, and the terms permit training on inputs. She reasons that planning applications are public documents, so nothing confidential is involved. Where is the gap in that reasoning?","options":[{"text":"Her own comments and questions go in alongside the public text, and those reveal her client's intentions.","correct":true},{"text":"Public documents may still be protected by copyright, so submitting them to a service that trains on inputs could infringe the rights of whoever prepared them.","correct":false,"tempting":"Raises a real and unsettled question about training and copyright, which does not bear on her client's confidentiality."},{"text":"Free tiers route to smaller models, so summaries of long planning documents will be less reliable and may omit conditions that matter to the client's decision.","correct":false,"tempting":"A sound reason to doubt the output quality, and it is an accuracy concern rather than the confidentiality gap in her reasoning."},{"text":"Planning documents contain the names and addresses of objectors and neighbours, which are personal data regardless of the document being publicly available on a council website.","correct":false,"tempting":"A genuine data-protection point worth acting on, and it concerns third parties rather than the client information her own prompts disclose."}]},"prev":{"slug":"the-bill-that-creeps","title":"The bill that creeps, and the quarterly hour that stops it","url":"https://addaly.com/learn/ai-for-small-business/the-bill-that-creeps"},"next":{"slug":"customer-data-and-the-law","title":"Your Customers' Data, and the Law Where You Are","url":"https://addaly.com/learn/ai-for-small-business/customer-data-and-the-law"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Free tiers, and what they take instead of money', in AI for a Small Business. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-small-business/free-and-what-it-costs (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Free tiers, and what they take instead of money","courseTitle":"AI for a Small Business","module":{"index":9,"slug":"cost-risk-and-the-long-run","title":"What it costs, what it risks, and what happens in a year"},"answer":null,"keyIdea":"A free tier is paid for in data rights, rate limits or a smaller model, and which of the three it is determines whether it is fine for your marketing copy and unacceptable for your client files.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"free-is-a-price-not-an-absence-of-one","runs":[{"text":"Free is a price, not an absence of one"}]},{"t":"p","runs":[{"text":"Every free AI tier is funded somehow, and for a small business the useful question is not whether it is free but what it takes instead."}]},{"t":"p","runs":[{"text":"There are only three answers, and they have very different consequences."}]},{"t":"h2","id":"one-your-inputs-train-the-model","runs":[{"text":"One: your inputs train the model"}]},{"t":"p","runs":[{"text":"Consumer free tiers frequently reserve the right to use what you type to improve the service. Paid business and API tiers usually do not, and say so in the terms."}]},{"t":"p","runs":[{"text":"Two things follow. This is a real distinction and it is usually the main thing you get for the money on a business plan. And it is a setting as often as a plan: several products let you turn training off on a free account, in a preferences screen nobody visits."}]},{"t":"p","runs":[{"text":"What it means practically: a customer’s medical detail, a client’s unannounced product, a supplier’s confidential pricing, a colleague’s grievance. Those do not go into a tier whose terms permit training, and the mitigation from the data lesson applies: strip identifying detail before pasting, and the question becomes much less sharp."}]},{"t":"p","runs":[{"text":"Check the terms, note the date you checked, and put that note beside the tool in your list. Terms change, and your obligation to your customers does not change with them."}]},{"t":"h2","id":"two-rate-limits-and-quiet-downgrades","runs":[{"text":"Two: rate limits and quiet downgrades"}]},{"t":"p","runs":[{"text":"The second currency is capacity. Free tiers cap messages per hour, files per day, or length. Fine for occasional work, and it fails exactly when you are busy, which is when you were relying on it."}]},{"t":"p","runs":[{"text":"Less obviously, free tiers are often routed to smaller models, and during heavy demand some products silently route everyone to a smaller one. You will not be told. The symptom is a day when the output is noticeably flatter and nothing you changed explains it, which is the argument, from the logging lesson, for recording which model produced what."}]},{"t":"h2","id":"three-the-free-tier-is-the-advert","runs":[{"text":"Three: the free tier is the advert"}]},{"t":"p","runs":[{"text":"The third kind is a genuine sample designed to get you onto a paid plan. Nothing sinister, and it has one operational consequence: free tiers get withdrawn or reduced with little notice, so anything your business depends on daily should not sit on one."}]},{"t":"h2","id":"the-free-path","runs":[{"text":"The free path"}],"role":"free"},{"t":"p","runs":[{"text":"Distinct from free tiers, and worth knowing because it is the only version that nobody can withdraw."}]},{"t":"p","runs":[{"text":"Open-weight models run locally.","b":true},{"text":" Ollama or llama.cpp on your own machine, with a model like Llama, Mistral, Qwen or Gemma. A recent laptop with 16GB of memory runs a small model at usable speed. No per-use cost, no rate limit, and (this is the real argument) nothing leaves the building, which makes the confidentiality question disappear rather than being managed."}]},{"t":"p","runs":[{"text":"The honest limitations: a small local model is meaningfully weaker than a frontier model at reasoning and long documents, setup is an afternoon, and the electricity is not zero. Module one covered this door; here it is the answer to a specific question, which is what to do when the data is the constraint."}]},{"t":"p","runs":[{"text":"Free tools that are not AI at all.","b":true},{"text":" LibreOffice, GIMP, Inkscape, Audacity, DaVinci Resolve’s free edition, Python. Much of what gets bought as an AI subscription is a task these already do."}]},{"t":"p","runs":[{"text":"Open-weight models on somebody else’s hardware.","b":true},{"text":" A middle option that gets overlooked: several providers host open models at a fraction of frontier prices, charged per token. You get the low cost and the ability to move between providers, because the model is the same wherever it runs, which is the lock-in argument from later in this module, arriving early."}]},{"t":"h2","id":"reading-the-terms-in-five-minutes","runs":[{"text":"Reading the terms in five minutes"}]},{"t":"p","runs":[{"text":"You are not reading the whole document. You are looking for four things, and a search box finds all of them."}]},{"t":"p","runs":[{"text":"Search for "},{"text":"train","b":true},{"text":". This tells you whether inputs may be used to improve the service, and whether there is an opt-out."}]},{"t":"p","runs":[{"text":"Search for "},{"text":"retain","b":true},{"text":" or "},{"text":"delete","b":true},{"text":". How long do they keep what you send, and can you make them remove it."}]},{"t":"p","runs":[{"text":"Search for "},{"text":"sub-processor","b":true},{"text":". Who else touches the data, which matters for the contracts lesson later in this module."}]},{"t":"p","runs":[{"text":"Search for "},{"text":"region","b":true},{"text":" or "},{"text":"transfer","b":true},{"text":". Where the processing happens, which matters in several regimes."}]},{"t":"p","runs":[{"text":"Write the four answers and the date on one line in your tools list. It takes five minutes per tool, once, and it converts a vague anxiety into a fact you can act on."}]},{"t":"h2","id":"choosing-in-one-line","runs":[{"text":"Choosing, in one line"}]},{"t":"p","runs":[{"text":"Confidential client material: paid business tier with training off, or a local model. Ordinary business writing and analysis: a free tier is fine. Anything the business depends on daily: pay, because dependence on something nobody owes you is a risk with no upside."}]}],"check":{"question":[{"text":"An architect uses a free chat tier to summarise planning documents, and the terms permit training on inputs. She reasons that planning applications are public documents, so nothing confidential is involved. Where is the gap in that reasoning?"}],"options":[{"text":[{"text":"Her own comments and questions go in alongside the public text, and those reveal her client’s intentions."}],"correct":true,"tempting":null},{"text":[{"text":"Public documents may still be protected by copyright, so submitting them to a service that trains on inputs could infringe the rights of whoever prepared them."}],"correct":false,"tempting":[{"text":"Raises a real and unsettled question about training and copyright, which does not bear on her client’s confidentiality."}]},{"text":[{"text":"Free tiers route to smaller models, so summaries of long planning documents will be less reliable and may omit conditions that matter to the client’s decision."}],"correct":false,"tempting":[{"text":"A sound reason to doubt the output quality, and it is an accuracy concern rather than the confidentiality gap in her reasoning."}]},{"text":[{"text":"Planning documents contain the names and addresses of objectors and neighbours, which are personal data regardless of the document being publicly available on a council website."}],"correct":false,"tempting":[{"text":"A genuine data-protection point worth acting on, and it concerns third parties rather than the client information her own prompts disclose."}]}],"order":[2,3,0,1],"why":null},"prev":{"slug":"the-bill-that-creeps","title":"The bill that creeps, and the quarterly hour that stops it","minutes":8},"next":{"slug":"customer-data-and-the-law","title":"Your Customers’ Data, and the Law Where You Are","minutes":9}}}