Will the next AI model fix your pitches?

No. A newer model writes more fluently, but it drafts from what the firm gives it: the last pitch, the team’s prompts and the files it can reach. It doesn’t know what this client cares about most or what will persuade them. That is the firm’s method, and a new model leaves it as good as you left it.

What does a new model change?

Mostly the writing. Newer models generally write more smoothly, follow instructions more closely and handle longer documents. If clumsy prose was holding your drafts back, a new model helps.

It also changes how the model reads your prompts. OpenAI’s developer guide to GPT-5.5 tells teams to “[b]egin migration with a fresh baseline”, because older prompts “often over-specify the process” and can lead to “overly mechanical answers”. Its guide to the newer GPT-6 models warns that they can be “more sensitive to instructions” in the files they read. So prompts tuned for an older model can give worse answers on a newer one. Retest them each time the model changes.

What won’t a new model change?

What the AI starts from. It drafts from your last pitch and whatever files it can reach, and a better model writes that material up better. If the material talks about the firm rather than the client, so does the better draft.

There is a public sample of how firms write about themselves. Taleist used AI to code every sentence on the homepages of 55 law firms from the Australian Financial Review’s partnership survey, captured on 25 July 2026, then checked the counts. Of the 738 sentences that put either the firm or the client in the frame, 694 were about the firm: 94%. None of the 55 opened with a line about the client. A homepage isn’t a pitch. But if a firm’s pitches share that habit, an AI fed those pitches copies it.

Nor will a new model supply the judgement: which requirement the client or panel ranks highest, and what this client needs to believe before choosing a firm it hasn’t used. That judgement is method, and someone has to build it in.

Is prompt engineering training enough?

No. It teaches the team to write clearer instructions, which is worth having, but a prompt can only be as good as its author. It writes to the standard of the person who wrote it, on the day they wrote it. Have whoever pitches best write the prompts, and retest them when the model changes.

What does context engineering add?

Your facts. Context engineering loads the AI with the firm’s documents: credentials, matter or engagement lists, past pitches and bios. It doesn’t add the argument, because nothing in those documents says how this client decides or how to persuade them. Add the method as its own layer: what clients think about when they choose, how a pitch for them should run, and examples that open on the client.

Why does a firm’s prompt library stop working?

Because each prompt was written by one person, for one model and for the old workflow, and all three move on. The author’s standard becomes the library’s ceiling. The model underneath changes. And the prompts carry the old way of pitching forward.

Give the library an owner. Keep a small set of test pitches and rerun them whenever the model changes. Keep the prompts that still produce a pitch that opens on the client, and rewrite the rest.

What fixes a pitch?

The four layers above the model: a workflow that starts each pitch from the client instead of the last pitch, a method for working out what this client values and how to persuade them, the firm’s knowledge, and instructions written to that method. The workflow, method and knowledge carry over to each new model. The instructions get retested. Then each new model writes the better pitch faster.

Taleist builds these layers, and calls the result consulting-grade AI: an advisor built on the firm’s knowledge and the craft of persuasion, running inside the AI the firm has already approved. Searchlight Group, a recruitment firm, pitches to government and not-for-profit evaluation committees, which score proposals the way a client panel does. It won 35% of its proposals before the work and 85% in 2025, and each proposal now takes two hours instead of six. Its managing director, Michael Holdway, puts the change down to “the strategic thinking behind every section”, not the writing.

It is the wrong route for a firm whose pitches already win and whose problem is speed. Pitch software fits that job better. Our guide compares the options: who builds AI for business development. To show the partners what the AI is winning, see how to show the return on AI in business development.