// guide
Why is "which LLM do you use?" the wrong question?
Updated: 2026-08-03
// short answer
The model inside an AI tool is not what determines the outcome. The harness around it — system prompt, tools, context, memory and working loop — turns the same LLM into completely different products. That same logic applies to how ChatGPT, Claude and Perplexity discover and cite your website, which is why GEO optimizes for the harness of generative systems rather than for a ranking algorithm.
One brain, different professions
When a business evaluates an AI tool, the conversation almost always starts with the model. "Is it GPT or Claude?" "Which version?" "The new model is smarter, so the product must be better, right?" It is an understandable question — and about as useful as choosing a contractor by the brand of their laptop.
Take the same model — say, Anthropic's flagship — and look at it inside two different products. The first is Cursor, a coding agent. Mention it in Slack with a task, and it spins up a virtual machine, clones your repository, writes code, runs tests, and delivers a pull request. A brilliant engineer.
The second is Claude Tag, a "general-purpose teammate" in that same Slack. It reads channel history, remembers project context, connects to your documents and calendars. It can write code too — but it will just as comfortably review a client thread and suggest how to answer a third request for a discount.
Same model inside. Two completely different specialists outside. Ask the coding agent about client communication strategy and it will answer something — the model is smart, after all. But it will be a plumber's take on living-room design: a capable person who simply showed up with different tools, in a different role.
What a harness is
The harness is everything that surrounds a model and turns it from a brain in a jar into a working tool.
- The system prompt — instructions that define the role, tone, priorities and boundaries. It decides whether an agent reduces every task to code or reasons about the business.
- Tools — what the agent can access. A terminal and git make it a programmer. A CRM and email make it a sales assistant. Web search and a crawler make it a researcher that reads your site and your competitors' sites before giving advice.
- Context and memory — what the agent knows about you. Channel history, company documents, past decisions. A model without context meets you for the first time in every conversation.
- The working loop — how the cycle is structured: a single reply, a long autonomous session, self-verification, escalation to a human.
Change the harness, change the product
Change the harness, and the same model becomes a different product. Leave the harness weak, and upgrading the model changes almost nothing.
Why this concerns your website
All of this applies not only to the tools you buy, but to how AI sees your business. When a potential customer asks ChatGPT, Claude or Perplexity "which service should I choose for X," the model does not answer from memory. It works inside its own harness: a search tool finds pages, a parser extracts text, and only then does the model compose an answer. At every step, your website either passes the filter or it does not.
A beautiful site built on heavy client-side JavaScript may be invisible to the parser. A page without clear structure — headings, facts, specifics — will be read but never cited: there is nothing for the model to lift. No mentions on external sources means the search step simply will not find you.
This is exactly what GEO (Generative Engine Optimization) is about: optimizing not for a ranking algorithm, but for the harness of generative systems — for how AI agents discover, read and retell your content. Classic SEO asked "how do we get into the top 10 links?" GEO asks "how do we get into the answer itself?"
What to do with this
If you are choosing an AI tool — look past the model and check how well the harness matches your task. A tool running a "lesser" model with the right instruments and access to your data will beat a flagship model in a generic chat window.
If you are building AI inside your company — start not with model selection but with harness design: what role the agent plays, what it sees, what it can do, where its boundaries are. The model slots in later and can be swapped almost painlessly.
And if you want AI systems to recommend your business — think of your website as part of someone else's harness. Your pages will become someone's context. The only question is whether they will be fit for it.
Related questions
What is a harness in AI products?
The harness is everything wrapped around a language model that turns it into a working tool: the system prompt, the tools it can call, the context and memory it has about you, and the working loop that structures how it acts. Change the harness and the same model becomes a different product.
Does upgrading the model fix a weak product?
Usually not. A weak harness — sparse tools, no memory, a generic prompt — caps what even a flagship model can do. A more modest model with the right instruments and access to your data will often outperform a stronger model sitting in a generic chat window.
How does the harness idea relate to GEO?
When ChatGPT, Claude or Perplexity answers a commercial question, the model works inside its own harness: search finds pages, a parser extracts text, then the model composes the answer. GEO optimizes your site for that harness — so agents can discover, read and cite you — rather than for a classic ranking algorithm.
Should I ignore which LLM a tool uses?
No — the model still matters. Treat it as one input among several. Ask first whether the harness matches your task: role, tools, access to your data, and how the working loop is designed. The model can often be swapped later; a mismatched harness cannot.