Free LLM Routing: What Each AI Tool Is Actually Built For

The short answer: open Perplexity when you need sourced facts, Claude when you need something written in your voice, ChatGPT when you need twenty options, Gemini when the work already lives in Google Docs, Grok when you need to know what people said in the last hour, and DeepSeek when the task is technical and the budget is thin. A free Perplexity account plus a free Claude account covers most weeks. Add ChatGPT Plus at $20 a month when you need volume.

The habit that costs you quality

Most marketers pick one assistant and run everything through it. Usually ChatGPT, because it was first and it is familiar.

That habit is the equivalent of cooking every meal in one pan. It works well enough, which is exactly why it survives, and it also means you are getting a mediocre version of half the tasks you do without ever seeing the better version to compare against.

The models are not interchangeable. They were trained differently, they are tuned for different behaviour, and the gap shows up most on the work that matters. Long-form writing in a brand voice is a genuinely different task from generating twenty headline options, and the tool that is good at one is usually not the best at the other.

What follows is how I actually route work across six of them, and what each is worth.

Claude: long-form and voice

Reach for Claude when the output is going to be read by a person and the wording matters. Long-form articles, email sequences, anything that has to sound like your brand rather than like software.

The specific strength is instruction-following over long spans. If you give Claude four paragraphs of context about tone, audience and what to avoid, it will still be honouring paragraph three by the time it writes section five. Most assistants drift.

The practical pattern: give it the brand context once, in detail, then ask for the piece. A prompt like "our voice is friendly, practical and empowering; the product saves solopreneurs time; avoid hype; give real value in every email" changes the output far more than any amount of rewording the instruction itself.

When not to use it: rapid divergent ideation. Ask for twenty variations and you will get a considered eight.

ChatGPT: volume and variation

Reach for ChatGPT when you want quantity and range. Twenty ad headlines, thirty subject lines, fifteen campaign angles.

It is unusually good at producing genuinely different options rather than twenty rewordings of the same idea, and it handles format switching well, so you can move from a list to a table to a script without re-explaining the task.

The honest use is as a starting spread, not a finished asset. Generate the twenty, keep the three that are actually distinct, then take those somewhere else to be written properly.

When not to use it: anything where a factual claim carries weight, unless you are checking it elsewhere.

Gemini: work that already lives in Google

Reach for Gemini when the document is already in Google Docs, the data is already in Sheets, and moving it out is the friction you are trying to avoid.

The advantage is proximity rather than raw capability. Summarising a strategy doc, drafting a content calendar from goals already written down, pulling live web results into a document you are editing: these are faster when the assistant is inside the tool instead of in another tab.

It also handles multimodal input, so images, audio and video go in alongside text.

When not to use it: as your writing tool, if your brand voice is distinctive. Proximity is convenience, and convenience is not the same as fit.

Perplexity: facts you can cite

Reach for Perplexity when you need something you can defend in a meeting.

It answers with numbered sources attached, which changes the workflow. Instead of getting a confident paragraph and then verifying it, you get the paragraph and the citations together and you click through to the ones that matter. For competitive monitoring, market sizing or anything with a statistic in it, this is the difference between research and a plausible guess.

This is also the habit that improves your own content. If you cannot find a source through Perplexity, you probably should not publish the claim.

When not to use it: drafting. It is built to answer, not to write at length.

Grok: what people are saying right now

Reach for Grok when recency is the whole point. It has live access to X, so it is useful for tracking how a platform announcement is actually landing, watching sentiment on a launch, or catching a trend while it is still a trend.

Treat the output as signal about a conversation, not as a measurement of a market. X is one platform with one demographic skew, and confusing the two is how people end up building campaigns for an audience that exists mainly on X.

When not to use it: anything where you need a representative sample.

DeepSeek: technical work on a budget

Reach for DeepSeek when the task is technical and the cost matters. Explaining an API call, writing a tracking script, debugging a tag implementation, working through a data transformation.

For a marketing team without an engineer, this is the tool that makes small technical tasks survivable. It is strong on code and cheap to run, which is a combination that matters when the work is high-volume and low-glamour.

When not to use it: anything customer-facing that needs polish.

The stack that covers most weeks

For most marketers, three accounts:

Tool	Cost	Job
Perplexity	Free	Research and fact-checking
Claude	Free	Writing and brand voice
ChatGPT Plus	$20/month	Ideation and deep research

The two free accounts do the majority of the work. The paid one earns its money when you are producing at volume, and until then the free tier is genuinely usable.

What I would not do is pay for four subscriptions. The marginal value of the fourth tool is low, and the real cost is not the money. It is the cognitive load of remembering which tab does what.

How to make routing a habit

The routing is easy to agree with and hard to actually do, because opening the familiar tab takes no thought.

Two things that make it stick:

1. Name the job before you open anything. Say out loud, or write down, what you actually need: sourced facts, a finished draft, options, live sentiment. The tool follows from the job. Most bad routing happens because the job was never named.
2. Keep the brand context somewhere reusable. The reason people stay in one tool is that re-explaining the brand is tedious. Write the context once, keep it in a note, paste it in. Switching becomes cheap the moment you stop retyping.

The honest summary

There is no best AI assistant, and the question is malformed. There is a best assistant for sourced research, a different one for long-form voice, and a third for volume.

The teams getting real leverage are not the ones with the most sophisticated prompts. They are the ones who stopped asking one tool to do six jobs.

Sources

The routing in this article is our own working practice, not benchmark data. Model behaviour changes with every release, and no published benchmark maps cleanly onto "which one writes in a brand voice." Treat the assignments as a starting point to test against your own work, and check each product page for current free-tier limits.

• Claude
• ChatGPT
• Google Gemini
• Perplexity
• Grok
• DeepSeek