Prepare a talk: what GPT-5.2 changes for speakers

To prepare a talk has meant something new since 11 December 2025: an AI model can draft the first set of slides in a few minutes, while the core message, the timing and the rehearsal remain with you. On that day OpenAI released GPT-5.2, which according to The Decoder draws level with professionals or beats them in 70.9 per cent of the comparisons in the GDPval benchmark for knowledge work, and according to InfoWorld OpenAI also describes it as better at building presentations. For speakers this shortens the time spent on the slide deck, while the work on the audience and on the rehearsal stays the same. In a survey by Talent Bureau of 250 working professionals in October 2025, 72 per cent prefer a keynote of 20 to 40 minutes, and 44 per cent stay most engaged by stories.

What OpenAI released with GPT-5.2 on 11 December 2025

OpenAI released GPT-5.2 on 11 December 2025. According to The Decoder the same day, OpenAI calls the series its most capable model yet for professional knowledge work, and it is rolling out first to paying ChatGPT users, including the Plus and Pro plans. The central figure comes from GDPval, a test of knowledge-work tasks in which assessors compare the model’s results with those of professionals: GPT-5.2 Thinking wins or ties in 70.9 per cent of the comparisons. OpenAI developed the test itself, so the figure is a claim by the manufacturer.

Another part of the news matters more to speakers. InfoWorld reported on 12 December 2025 that OpenAI describes the model as better at building presentations and shows a fully formatted workforce-planning spreadsheet as an example, which the previous model had delivered with the right content but without formatting. The report gives no figure for the quality of presentations. All that can be documented is therefore a direction: a draft with an outline and a layout costs less and less time.

Prepare a talk: what an AI draft delivers and what stays with you

To prepare a talk means all the steps between an organiser’s acceptance and your appearance at the conference. First you clarify the audience and the core message, then come the structure of the session, the building of the slides and the rehearsal. An AI model can help with each of these steps, but the parts that convince a specialist audience need knowledge that only you have.

The table reflects our assessment and follows from a simple observation: a model knows neither the room nor your projects, and what you do not give it, it fills in from what already appears on many conference slides. A draft built from general material therefore resembles the talks the audience has already heard.

What conference audiences say they want from a talk

When you prepare a talk, the question of the audience matters most, and a recent figure on it comes from the Talent Bureau survey, published on 30 October 2025. It asked 250 working professionals, for example in marketing, events or sales, who had attended a work-related event in the previous twelve months, and the fieldwork ran in October 2025.

According to the published results, 72 per cent of respondents prefer keynotes of between 20 and 40 minutes, and more than 58 per cent prefer sessions with some interactivity, such as Q&As or polls. Respondents stay most engaged by stories (44 per cent), practical how-to lessons (37 per cent) and humour (32 per cent). As preferred formats they name interactive workshops (44 per cent), panel discussions (40 per cent) and keynotes (31 per cent). For planning a talk this means that a 30-minute slot sits in the middle of the range most respondents want, and that a concrete case from your own work holds attention better than an overview. Which lengths conferences actually award is covered in the article How long is a conference talk?.

The survey comes from a single provider and rests on 250 respondents, so it works as a guide for designing a talk.

Time plan: how many hours it takes to prepare a talk

Because few people ever work out the effort, here is a calculation with openly stated assumptions. Suppose you prepare a 30-minute specialist talk and set aside 20 hours: three for the audience and core message, three for the structure, six for the slides, five for the rehearsal and three for technology and a buffer. If you give six talks a year, that is 120 hours. If an AI draft saves half the slide time, which is three hours per talk, that makes 18 hours a year that you can put into rehearsal. All the figures are assumptions that you adjust to your own experience, and what stands out in the distribution is the rehearsal, which at five hours takes almost as much time as the slides and cannot be handed to a model.

How to prepare a talk in six steps

1. Clarify the audience: read the conference programme, ask the organiser about the role of the session and note three questions your audience should have answered after the talk.
2. Write the core message in one sentence: the sentence names a result you can back up from your own work, with a figure or a case. How session types differ is explained in What formats are there at a conference?.
3. Have an outline suggested: give an AI model the audience and the core message as well as the slot length, and check the proposal against your three questions from step 1.
4. Build the slides with your own material: a draft from the model serves as a frame, the figures and examples come from you, and each slide carries one statement.
5. Rehearse three times aloud: in the first run you time yourself, in the second you cut, in the third you speak to a colleague who may ask questions.
6. Plan technology and a buffer: test your laptop and adapter before the talk and keep the closing slide ready in case time runs short. What happens at a conference before you go on is described in the article on the keynote at a conference.