Using AI to support the hobby — in practice
The basics article covered where AI is suited and where its limits lie. This article gets practical: how to start using the tool, how to set it up for a hobbyist’s needs, and what kinds of prompts get the most out of it. Claude is used as the example, because it is the one most familiar to us, but the same principles apply to other language models.
Claude in a nutshell
Claude is a language model developed by Anthropic that can be used in a browser at claude.ai as well as with iOS and Android apps and a desktop app. Using it requires no technical skill — the conversation flows much like messaging.
The free version goes a long way: it includes handling of text, images and code as well as web search, and works on all devices. Free use is, however, limited to a certain number of messages within a given period, and it uses lighter models than the paid versions.
The paid Claude Pro (about €22/month) raises usage limits clearly, gives access to the most capable models, and unlocks a feature especially useful to a hobbyist: projects. Pro also includes the Cowork tool, which runs in the desktop app (currently on macOS, in research preview) and whose special trait is that it can work directly with files on your computer — for example, open and update a water-parameter tracking spreadsheet. The more expensive Max tiers offer the same content but with even higher usage limits — they are aimed at very active users. For most hobbyists the free version or Pro is plenty.
The best basic setup for a hobbyist
You get considerably more out of AI when you give it a persistent context about your own tank instead of repeating the basics in every conversation. Claude’s projects are an excellent tool for this.
A project is its own conversation space to which you can attach persistent background instructions and files. In practice a hobbyist can create a project named, say, “Reef aquarium” and give it:
Persistent background instructions. A short description of the tank: size and water volume, age, fish and corals, the salt used, lighting and flow, and the dosing method. This way Claude knows the starting point every time without a separate explanation.
A knowledge base as files. You can attach, for example, a water-parameter tracking spreadsheet, a device list, or manufacturer instructions in PDF form. Claude can use these in its answers, so the interpretation is based on your own data.
Your own response style. You can instruct Claude to answer the way you like — concisely, citing sources, or always having dosages double-checked separately. The instruction stays in force for the whole project.
With this setup the conversation changes fundamentally: you are no longer asking general questions of a general assistant, but talking with an assistant that knows precisely your tank.
The structure of a good prompt
A prompt is the message you write to the AI. The quality of the answer depends directly on the quality of the prompt. A good prompt typically has four parts: context (which tank and situation it concerns), data (readings, images, observations), the question (what you want to know) and the format (how you want the answer).
Weak prompt: “Why is the coral doing poorly?”
Good prompt: “The Euphyllia in my mixed reef tank (300 l, 14 months old) has been retracted for two days. Alkalinity 7.8 dKH, calcium 420, magnesium 1300, salinity 1.025, nitrate 5, phosphate 0.03. Nothing has been changed except I just replaced the skimmer cup. What might cause the retraction, and in what order should I rule causes out?”
The latter gives Claude everything it needs and asks for a structured answer. The more precisely you describe the situation, the more accurate the answer.
Practical examples
Interpreting water parameters. Enter your latest measurements — preferably from several measuring sessions — and ask Claude to assess the balance and trends. You can ask, for example, whether some value is drifting out of the target range and what that says about consumption. If you have a tracking spreadsheet in the project, Claude can examine the development over a longer period. The same applies to broader ICP analyses: you can feed the result to Claude and ask it to spell out which values need attention.
More on this topic: Laboratory tests (ICP).
Identifying a problem from an image. Attach a photo of algae, a bleached coral or a suspicious organism and describe the situation briefly. Claude can suggest what it is and ask follow-up questions to refine the identification. Remember that image analysis is not certain — use it as a starting point, not a final diagnosis.
More on this topic: Dinoflagellates — in practice and Cyanobacteria — in practice.
Calculating a dose. When you want to raise, say, alkalinity by a certain amount, you can ask Claude to calculate the dose based on your water volume and the product you use. Always check the calculation against the manufacturer’s instructions — this is exactly the kind of figure AI can get right but can also get wrong.
More on this topic: Dosing methods.
Care plan and troubleshooting. In a multi-faceted problem Claude is at its best as a conversation partner: you can describe the situation and ask it to draw up a step-by-step plan in which causes are ruled out in order. This way you avoid hasty fix attempts that change several things at once — a common mistake in a reef aquarium.
Other language models worth considering
Although the focus here was on Claude, a hobbyist does fine with other tools too. Based on the available information:
ChatGPT (OpenAI) is the most widely used and very versatile. It is strong at writing and multi-step reasoning, and a wealth of plugins is available for it. The principles — project-like customizations, image analysis and good prompts — work much as they do with Claude.
Gemini (Google) stands out for its ability to handle very large amounts of information at once, which is useful if you want to feed it long documents or extensive tracking data. It is also tightly connected to Google’s services, such as Sheets and Drive.
The main thing is not to choose the “right” model but to learn to use one tool well. The principles of a good prompt and of your own tank’s context apply to all of them.
Pitfalls and a checklist
Finally, a few rules of thumb so that AI is a help and not a hindrance:
- Check the figures. Dosages and other critical calculations are always verified against the manufacturer’s instructions.
- Give enough context. The more precisely you describe the tank and the situation, the better the answer.
- Don’t change many things at once. Ask for a plan that proceeds in a controlled way.
- Stay critical. If an answer feels odd or contradicts what you knew before, ask for the reasoning and check your sources.
- Remember that AI does not measure for you. An answer is only as good as the data given to it.
Used correctly, AI is a hobbyist’s best study companion — it speeds up learning, helps you grasp the big picture and acts as a patient adviser. The final responsibility for the tank still always stays with you.
More on this topic: Using AI to support the hobby — the basics.