FA
Quick Answer:AI assistants don't cite stages — they cite documents, and mostly fresh ones: 76.4% of ChatGPT citations point to content updated within the previous 30 days (ConvertMate, January 2026). The stage-to-citation loop converts each talk into citable assets in five steps: a question harvest captured on stage and off it, a question-shaped article carrying the slide data within two weeks, a quotable organizer recap that earns a trusted third-party mention, the recording published under a claim-naming title, and a quarterly refresh that keeps the asset inside the recency window.
The 40-minute lifespan problem
Here is the uncomfortable arithmetic of speaking. A keynote reaches the people in the room, for the duration of the slot, once. Meanwhile the surface where your next booking decision gets researched is growing at industrial speed: AI search drove an estimated 27.4 billion visits in Q1 2026, up 42.8% year over year, per Wix Studio's AI Search Lab analysis of Similarweb data. As I showed in GEO for Speakers, the shortlist for your next stage is increasingly assembled inside a model — and the model was not in the room.
Generative engines cite documents: pages they can retrieve, parse, and quote. A talk that never becomes a document is, from the engine's point of view, an event that never happened. And even documents age out fast — ConvertMate's January 2026 analysis of 10,000+ domains found 76.4% of ChatGPT citations pointed to content updated within the previous 30 days. The conclusion I build everything on: the pipeline, not the podium, earns the citations.
The stage-to-citation loop at a glance
I sketched this loop in one paragraph of the GEO for Speakers post; this is the full specification. Five steps, in strict order, with deadlines attached:
- Step 1 — Question harvest, during the event and the days after. Captured at the mic, in the event app, and in every queue and dinner around the stage.
- Step 2 — Article, within two weeks. The core argument rebuilt as a question-shaped, sourced document carrying the slide data.
- Step 3 — Organizer recap, same two weeks. A quotable summary sent to the event team — the cheapest earned mention in GEO.
- Step 4 — Recording, when the event releases one. Published under a claim-naming title, not an event-naming one.
- Step 5 — Refresh, quarterly. Substantive updates that keep the asset inside the recency window.
One talk in, five citable surfaces out. The rest of this post is the how, step by step.
Step 1 — The question harvest: on stage, off stage, and days after
I don't record my talks, and the pipeline doesn't need me to. The argument already exists — in the deck, in the sources, in the delivery notes. What exists nowhere else, and what no competitor can reproduce, is what the room did with the argument: the questions. Stage questions get logged verbatim within ten minutes of stepping down — the method from The Q&A Is the Research. But the mic captures only a fraction, because most people will not ask in front of five hundred peers.
So the harvest is engineered, not hoped for. The closing slide carries a QR code that connects directly to me — an open channel for the question someone wasn't ready to ask in public. The event app gets worked, not just installed: every DM and comment thread answered while the event is still running. And the physical spaces do the rest — the networking dinner, the coffee queue, the hallway between sessions. Everything around a conference is an excuse to interact, and the delayed questions that arrive through those channels — over dinner, or on LinkedIn three days later — are routinely more honest than the public ones: less performed, closer to the asker's real problem.
By the end of the event week, the harvest is one file: every question, verbatim, tagged by channel. That file — not a recording — is the ore the article in Step 2 is built from.
Step 2 — The two-week article, question-shaped
This is the step most speakers skip, and the one that matters most. The talk's core argument becomes a standalone article — not a recap. The distinction is structural:
- A recap says “I spoke at X, here are three takeaways.” It is about the event. Nobody asks an AI assistant about your event.
- An argument document makes the talk's claim in full, under headings shaped like the questions buyers actually ask, with every slide statistic carried over with its source and date. It is about the problem — and the problem is what gets prompted.
The formatting levers are the ones that measurably move retrieval: the 2023 Princeton and Georgia Tech GEO study found that adding statistics, citations, and quotations lifted a source's visibility in AI answers by up to 40% — and a data-dense keynote already contains all three; the article just has to preserve them. The FAQ block comes straight from the Q&A log, which is material your competitors cannot reproduce, and it earns the FAQ treatment I described in Your FAQ Page Is Your Most Underrated GEO Asset.
Why two weeks? Because the window is set by forces on both sides. Your recall of the room decays. And the engines skew recent — publishing while the event's recap pages, social posts, and attendee mentions are still fresh means your article enters the citation pool surrounded by corroborating signals, not as an orphan.
Step 3 — The organizer recap: the cheapest earned mention in GEO
Within the same two weeks, the organizer gets a gift: a short, quotable summary of the talk — the claim in two sentences, the two or three headline numbers with sources, a pull quote, and a link to the article from Step 2. Most event teams write recap pages anyway; you are simply making yours the easiest speaker section to write, and the only one with a ready-made quote.
The GEO logic: when someone prompts an assistant for speaker recommendations, the engines lean on third-party pages — agendas, recaps, roundups — not on speakers' own sites. Profound's analysis of 680 million citations found only 11% overlap between the domains ChatGPT cites and those Perplexity cites — you cannot predict which platform will serve your next organizer, so you want your name on trusted third-party surfaces across the board. An organizer recap that names you, quotes your claim, and links your article is precisely that surface, earned with one email.
Step 4 — The recording, named after the claim
When the event releases the recording — and most professional conferences do — the title decides whether it can ever be retrieved. “Fernando Angulo — Keynote — MarketingConf 2026” is invisible to every question-shaped prompt a buyer will ever type. “Why AI Visibility Doesn't Transfer Between Platforms” is a claim; claims get matched to questions. Google's AI surfaces draw on YouTube regularly, which makes the recording a first-class citable asset if — and only if — the metadata treats it like a document: claim-naming title, chaptered structure, and the key sources in the description.
The recording also compounds the other steps: the article embeds it, the organizer recap links it — and the ask costs nothing extra, because it travels in the same email that delivers the Step 3 recap to the event team.
Step 5 — The refresh cadence: recency is a ranking factor now
The 76.4% figure deserves to be read again, because it changes the economics of everything above: three-quarters of ChatGPT's citations go to content updated in the previous 30 days. A brilliant article published after a talk and never touched again will rotate out of the citation pool within a quarter or two, regardless of quality.
So the loop doesn't end at publication. Each pillar article gets a quarterly refresh — and a refresh means substance: the newest figures from the current version of the talk, a fresh example, an updated FAQ answer from the latest Q&A log. This is where speaking at volume becomes an unfair advantage: a speaker delivering the evolving version of a talk across 50+ stages a year generates genuine updates continuously. The refresh isn't a chore; it's a byproduct of the calendar. Rotating a date stamp without changing content, by contrast, is signal-gaming that platforms learn to discount — don't.
Closing the loop: measure it or it's a ritual
The pipeline produces assets; the Shortlist Test from GEO for Speakers measures whether they work: 8–10 organizer-style prompts, run across three platforms, mentions and citations scored separately, re-run monthly. When an article from Step 2 starts appearing as a cited source for the prompts your buyers actually type, the loop has closed — the stage produced the document, the document produced the citation, and the citation produces the next stage.
That is the whole system: five steps, one hard deadline, one quarterly rhythm. None of it requires a team — I run it from a notes file, a QR code on the last slide, and a calendar reminder. What it requires is treating the talk not as the product, but as the raw material of the product. The product is the citation.
Frequently Asked Questions
Don't summarize the talk — extract its core argument and rebuild it as a document: question-shaped headings that match how buyers phrase the problem, the slide statistics carried over with their sources and dates, and an FAQ block built from the real questions the audience asked. A recap (“I spoke at X, great crowd”) earns no citations; a self-contained, sourced argument does. Target publishing within two weeks of the talk.
Within two weeks. Your recall of the Q&A and hallway conversations decays fast, and generative engines skew heavily recent — ConvertMate found 76.4% of ChatGPT citations pointed to content updated within the previous 30 days. Publishing while the event's recap pages and social mentions are still fresh means your article enters the citation pool surrounded by corroborating signals.
Google's AI surfaces — AI Overviews and AI Mode — draw on YouTube regularly, which makes the recording a citable asset if the title names the claim rather than the event. “Keynote at MarketingConf 2026” is invisible to a question-shaped prompt; “Why AI Visibility Doesn't Transfer Between Platforms” can be retrieved. Add chapters and put the key sources in the description.
Quarterly is a working cadence for pillar posts, given the 76.4%-within-30-days recency skew. But a refresh only counts if it is substantive: updated figures, a new example, a revised FAQ answer from the latest Q&A. Rotating a date stamp without changing the content is signal-gaming that platforms learn to discount.
Yes, disproportionately. For speaker-recommendation prompts, engines lean on third-party pages — agendas, recaps, roundups — rather than the speaker's own site. Every organizer recap that names you, quotes your claim, and links your article is an earned mention on exactly the domain class the engines trust for that query. One quotable summary sent within days of the event is the cheapest link building a speaker can do.
Want the talk that feeds this pipeline on your stage?
I keynote 50+ events a year on AI search and GEO — and every talk ships with the quotable recap, the sourced data, and the article your recap page can cite.