The room and the record
A conference talk is two things that get sold as one. The first is an event: a few hundred people in a room, a stage, forty minutes, a queue afterwards. The second is a set of documents the event throws off — a listing page with a name and a headshot, a video, an auto-generated transcript, sometimes a paper, sometimes a slide deck on a file host.
The first of those is not addressable. Nothing indexes a room. Search engines, knowledge graphs and answer engines operate on documents with URLs, and a talk delivered to an audience produces no URL by the act of being delivered. If authority is understood as something a machine can observe, the talk contributes exactly as much as its artefacts and no more.
That sounds like a rhetorical framing, but it turns out to be the position taken by the vocabulary everyone uses to describe events on the web. It is worth looking at, because it is unusually explicit.
What the schema itself says about speakers
Schema.org's Event type is the standard machine description of a conference, a session or a talk. It carries dozens of properties: dates, locations, ticketing, attendance mode, organiser, audience. It does not carry a property called speaker.
The nearest thing is performer, defined as “A performer at the event—for example, a presenter, musician, musical group or actor”. A presenter is in that list, so a conference speaker can be described — but as a performer at an event, not as an author of anything.
Google's own Event structured data documentation is narrower still. The required properties there are name, startDate and location (with a nested postal address). performer appears in the recommended list, where Google glosses it as “The participants performing at the event, such as artists and comedians” and says to use “a nested PerformingGroup or Person, one for each performer”. A conference can publish a fully valid, rich-result-eligible event page that names no human being at all.
Now the interesting part. Schema.org does define a property for the thing that outlives an event, and it is not a person: recordedIn, “The CreativeWork that captured all or part of this Event”. Event also carries workFeatured and the general-purpose subjectOf. The vocabulary's model of how an event persists is that a separate creative work captures it and that work is the durable object. That is not an argument we are making about conferences; it is the data model, written down, by the people whose job was to describe events to machines.
The practical consequence is that a speaker page on a conference site is doing most of the work that anyone attributes to the talk. It is a URL with a name on it, a bio, usually a photograph, sometimes a link back to a company. It is a document, so it can be crawled, indexed and cited. The talk it describes cannot.
Recorded, indexed, transcribed: what each step costs
Assume the talk was filmed. That is now common but not universal, and the recording is where most of the durable value sits, so it is worth being precise about what has to happen before a video becomes something a search system can use.
Google's video documentation sets out the conditions plainly, and they are all about the page rather than the content. “The watch page must be indexed.” “The indexed watch page must be performing well in Search before its video can be considered for indexing.” The video must be embedded on that watch page, it “can't be hidden behind other elements”, and it “must have a valid thumbnail that's available at a stable URL”. Read that second condition again: video indexing is downstream of the page already doing well in ordinary search. A recording parked on a conference microsite that nobody links to is not obviously going to clear it.
The same document says nothing whatsoever about transcripts or captions. We checked for them specifically. They are not in Google's video indexing requirements, and as we found in does a YouTube presence affect AI answers, they are not in the VideoObject structured data reference either. That absence is not proof that spoken words are ignored — it is proof that there is no documented mechanism to point at.
Transcription does happen regardless. YouTube states that it “can use speech recognition technology to automatically create captions for your videos”, across roughly ninety languages, and is candid that the output is imperfect: automatic captions “might misrepresent the spoken content due to mispronunciations, accents, dialects, or background noise”, and creators should “always review automatic captions and edit any parts that haven't been properly transcribed”. YouTube also lists conditions under which they simply do not appear — unsupported language, poor sound quality, long silences, overlapping speakers. A panel with four people talking over each other in a hotel ballroom is close to the worst case the documentation describes.
The reliable move, and the boring one, is to publish the transcript yourself as text on a page you control. That converts a video into an indexable document with your name attached, on a domain whose survival you decide. We made the same argument about audio in whether transcripts matter more than the audio and does podcast audio get indexed; the mechanics are identical, because in both cases the machine-readable object is text on a URL.
What the published documentation requires at each step
Source: Google Search Central, Video SEO best practices and Top ways to ensure your content performs well in Google's AI experiences; YouTube Help, Use automatic captioning. All read 26 September 2026.
Bars encode position in the sequence, not magnitude or probability. Nobody publishes drop-off rates between these steps, and we are not inventing any.
Do conference sites keep the speaker page up?
This is the question that decides whether a speaking credit is an asset or a memory, and it is the question on which the evidence is weakest. There is no published measurement of conference website persistence. We looked, and if one exists we did not find it, so what follows is general web decay applied to a case that has reasons to be worse than average.
Pew Research Center published the best available general figures in May 2024. Sampling roughly 90,000 pages a year from 2013 to 2023 through Common Crawl — just under a million pages in total — and then trying to load them in October 2023, it found that “38% of webpages that existed in 2013 are not available today”, against 8% of pages that existed in 2023. Across the whole decade, “a quarter of all webpages that existed at one point between 2013 and 2023 are no longer accessible”. The breakdown matters for our purposes: 16% of pages were individually gone from a domain that still worked, and 9% were gone because the entire root domain had stopped functioning.
Per cent of pages, by the year the page was sampled
Source: Chapekis, A., Bestvater, S., Remy, E. and Rivero, G., When Online Content Disappears, Pew Research Center, 17 May 2024. Roughly 90,000 pages sampled per year from Common Crawl archives, just under one million pages in total, tested for accessibility as of October 2023.
These are figures for the web as a whole. They are not a measurement of conference websites, and we are using them as a floor for a category that plausibly decays faster, not as a substitute for evidence about it.
Why would conference pages do worse than the average page? Three structural reasons, none of which we can quantify. Many conferences run each edition on its own microsite, sometimes its own domain, which has no constituency once the event is over. Sponsorship and agenda pages are the first things dropped in a rebuild. And a conference that stops running, or an organiser that is acquired, takes its entire archive with it — the 9% root-domain failure mode, which is the one that destroys every page at once.
The counter-examples are real and instructive. Standards bodies, long-running academic conference series and organisations with an archival mandate keep everything, sometimes for decades, because permanence is part of what they are for. A talk given at a venue with an archive behind it is a materially different asset from the same talk given at a commercial summit in its second year. Nobody sells speaking slots on that distinction, and it may be the single most important variable.
The general problem of coverage outliving its usefulness is the subject of how long a placement stays findable and what happens when a publication shuts down. Conference pages belong in the same risk category, with less institutional weight behind them than a newspaper archive.
Can an assistant cite any of it?
Only if it became a page. Google's guidance for AI Overviews and AI Mode gives one condition and repeats that there is nothing else: to be shown as a supporting link, “a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements”, and “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary”. The same document describes a “query fan-out” technique, where the system issues multiple related searches across subtopics to build a response — which is still retrieval over indexed pages.
That is a rule about documents, and it is the whole answer to the question in this section. An appearance cannot be a supporting link. A speaker page, a recording's watch page, a transcript page or a proceedings entry can be, on exactly the same terms as any other URL. If you want the practical timing behind that condition, we cover it in how long Google takes to index a new page.
For assistants that are not Google, the picture is less documented but points the same way. The February 2026 evaluation of six commercial chatbots by Suzgun and colleagues — 2,100 questions put to six systems over a fortnight — found that retrieval failures, rather than reasoning failures, accounted for more than 70% of errors, and that the systems showed a marked pull towards a small set of well-known sources: models answering Hindi-language queries cited English Wikipedia more often than any Hindi outlet. That is a preprint and not peer-reviewed, and it was not designed to answer anything about conferences. We cite it for one narrow point: what gets into an answer is decided by a retrieval layer with strong preferences, and a conference microsite is not the kind of domain such a layer reaches for.
Where a talk does show up in this machinery, it is usually because someone else wrote about it. Coverage of what was announced on stage, a write-up on a trade site, a Wikipedia citation — those are third-party documents about the talk, and they behave like any other coverage. That route is covered in how an entity enters the knowledge graph and, for the narrower question of whether saying your name aloud accomplishes anything, in does saying your name on a podcast build an entity. The finding there transfers cleanly: being spoken about is not the same as being recorded in a document, and only the document is machine-legible.
A paid slot and a selected slot
Conference speaking is sold. Sponsorship prospectuses routinely package a session with a booth and a logo, and there is nothing improper about that as long as nobody pretends otherwise. The question for this guide is narrower: do the two kinds of slot leave behind different things?
Yes, in one respect that is documented, and probably in several that are not.
The documented difference is the proceedings. A slot won through a reviewed call for papers in an academic or standards-adjacent conference usually ends with a paper in published proceedings. Google Scholar's inclusion guidelines state that it indexes “journal papers, conference papers, technical reports, or their drafts, dissertations, pre-prints, post-prints, or abstracts”, and set out the technical conditions: files must be HTML or PDF, “PDF files must have searchable text”, every article URL must be “reachable from the homepage by following at most ten simple HTML links”, and the metadata must carry a title, the full name of at least the first author and a year of publication. A proceedings paper meeting those conditions enters a citation graph that other researchers can cite and that persists independently of the conference's own website.
Most published proceedings also carry a DOI. The DOI Foundation describes a DOI as “a digital identifier of an object, any object — physical, digital, or abstract”, functioning as a persistent identifier that provides “a long-lasting reference to an object or to information about an object”, resolvable through a proxy server. It is worth being precise about what that buys: the DOI Foundation is explicit that persistence rests on organisational commitment rather than on the technology, so a DOI is a promise by a publisher, not a physical guarantee. Still, a promise from a publisher with an archival mandate is a great deal more than a URL on a microsite. The broader question of what identifier registries do and do not achieve is covered in do identifier registries affect search visibility.
A sponsored slot rarely produces any of that. It produces an agenda listing, possibly a recording, and a relationship with the organiser. Those can be valuable — a good recording on a well-trafficked channel is a real asset — but nothing in the chain has been assessed by anyone other than the person who took the payment.
The undocumented differences are where caution is required. It is widely assumed that a reviewed slot reads as a stronger credential to a journalist, a Wikipedia editor or an immigration officer than a purchased one. We think that assumption is reasonable. We can point to no study measuring it, and we are not going to dress a reasonable assumption as a finding. What can be said with confidence is structural: selection creates an external record of a judgement, and payment creates an internal record of a transaction, and only the first is visible to anyone downstream.
One vigilance note. Between genuinely reviewed conferences and honestly commercial ones sits a category of events that solicit paid presentations while presenting themselves as scholarly. Think. Check. Attend., an initiative run by the publishing-services company Knowledge E, offers a three-step framework — think whether it is the right conference, check it against a list, then attend — along with a conference-checking tool. We flag it as a starting point rather than an authority: it is a checklist maintained by a commercial actor in the scholarly-publishing market, not a register, and we could not retrieve its full criteria to reproduce here. The same evaluative problem in a different market is covered in what business awards cost, where the pricing is at least published.
What a slot is actually worth, practically
Everything above points one way, so it is worth stating the operational version.
Treat the booking as the acquisition of publishing rights, not of stage time. Before agreeing, establish three things in writing: whether the session will be recorded, who owns and hosts the recording, and whether you may republish the video and its transcript on your own site. A conference that will not let you republish your own talk is selling you the room and keeping the artefact, which inverts the value.
Then publish. A page on your own domain with the talk title, an abstract, the slides and the full transcript is the object that gets indexed, survives the conference's next site rebuild, and satisfies the one condition Google states for AI citation. It is also the thing you can link to in a pitch three years later. The broader construction of that kind of owned footprint is the subject of the founder authority footprint.
And be honest in your own records about which kind of slot it was. A purchased keynote and a reviewed paper are both legitimate; they are not the same evidence, and anyone assessing the claim — a reporter, an editor, an adjudicator — will eventually ask which it was. The parallel question for written placements is covered in whether sponsored articles count as visa evidence, where the distinction turns out to matter a great deal.
Methodology and limitations
This guide is built from platform and vocabulary documentation rather than from measurement, because measurement of this question does not exist. The structured-data findings are from schema.org's Event type definition and Google Search Central's Event structured data reference, both read on 26 September 2026; the absence of a speaker property and the definitions of performer and recordedIn were checked directly against those pages. The video indexing conditions and the statement that transcripts do not appear in them are from Google's Video SEO best practices page, read the same day. The AI citation condition is from Google's Top ways to ensure your content performs well in Google's AI experiences. The captioning account is from YouTube Help. The proceedings and indexing account is from Google Scholar's published inclusion guidelines and the DOI Foundation's own description of what a DOI is. Five limitations should be read alongside the findings. First, and most important, we found no study measuring whether conference speaking changes any observable outcome — brand search volume, citation frequency in AI answers, knowledge panel presence, coverage volume — and the guide therefore describes mechanisms rather than effects. Second, the persistence figures are Pew's measurement of the web at large, not of conference sites; we have given reasons to expect conference pages to decay faster and no data confirming it, and readers should treat 25% as a floor rather than an estimate. Third, Google's documentation describes Google. Other search and answer systems publish far less, and the Suzgun preprint we cite for retrieval behaviour is not peer-reviewed and was not designed to study conferences. Fourth, conference practice varies enormously by field: an ACM or USENIX proceedings paper and a marketing summit keynote share a word and almost nothing else, and we could not read several publishers' proceedings policies directly because they refused automated access. Fifth, we could not retrieve the full Think. Check. Attend. checklist or a substantive report on predatory conference practices, so the pay-to-present section stops at the structural point and makes no quantitative claim about prevalence.
Conclusion
The honest answer is the one the data model already gives. Schema.org describes a conference as an event with a name, a date and a place, offers no way to say who spoke except by calling them a performer, and provides a dedicated property — recordedIn — for the creative work that captured it. Google's rules for indexing and for AI citation are rules about pages. What is durable about a talk is therefore whatever the talk produced in document form: a speaker page while the conference site lives, a recording whose watch page can clear Google's conditions, a transcript, a proceedings paper with an identifier behind it. The room is genuinely valuable for reasons that have nothing to do with search — the people in it, the questions afterwards, the relationships — and it should be valued for those. As an authority signal that a machine can see, a talk is worth precisely what you publish from it. Negotiate for the recording, publish the transcript on your own domain, and keep a clear record of whether the slot was bought or won.
Frequently Asked Questions
Does speaking at a conference help you rank or get cited?
Only through what the talk leaves behind. Google's documented eligibility rule for its AI features is that a page “must be indexed and eligible to be shown in Google Search with a snippet”. That is a rule about pages. A talk that produces no indexed page produces nothing a search or answer system can act on, however good the talk was.
Is there a speaker property in structured data?
No. Schema.org's Event type defines no property called speaker. The nearest is performer, defined as “A performer at the event—for example, a presenter, musician, musical group or actor”, and in Google's own Event structured data documentation performer is only recommended, not required, and is glossed as “artists and comedians”. The required properties are name, startDate and location.
What does schema.org say survives an event?
It has a property for exactly that. Event.recordedIn is defined as “The CreativeWork that captured all or part of this Event”. The vocabulary models the durable object as a separate creative work that points back at the event, which is a fair description of how authority from a talk actually travels.
Do conference websites keep speaker pages online?
No published study measures conference sites specifically. The best general evidence is Pew Research Center's May 2024 analysis of about a million sampled pages, which found 38% of webpages that existed in 2013 were no longer accessible in October 2023, and that a quarter of all pages sampled across 2013 to 2023 were gone. Conference sites carry the extra risk of per-year microsites and domains that lapse, but we have not seen that risk quantified.
Does a recorded talk get indexed automatically?
Not automatically. Google's video documentation sets conditions: “The watch page must be indexed”, “The indexed watch page must be performing well in Search before its video can be considered for indexing”, the video must be embedded and not hidden, and it must have a valid thumbnail at a stable URL. Transcripts and captions are not mentioned anywhere in that document.
Does a paid speaking slot signal the same thing as a selected one?
Not to anything that checks. A slot won through a reviewed call for papers usually ends in proceedings with a DOI and an entry that Google Scholar will index, because Scholar accepts “journal papers, conference papers, technical reports, or their drafts”. A sponsored slot typically ends in a listing page and, at best, a video. Both can produce a good artefact; only one produces a record that a third party has vouched for.
Sources
- vocabulary Schema.org, Event. Property list; definitions of performer and recordedIn; confirmation that no speaker property exists. Read 26 September 2026.
- documentation Google Search Central, Event (Event, Place, Offer) structured data. Required properties name, startDate, location; performer as a recommended property. Read 26 September 2026.
- documentation Google Search Central, Video SEO best practices. Video indexing requirements, including the watch-page and Search-performance conditions; key moments. Read 26 September 2026.
- documentation Google Search Central, Top ways to ensure your content performs well in Google's AI experiences. AI Overviews and AI Mode eligibility rule; query fan-out. Read 26 September 2026.
- documentation Google Scholar, Inclusion guidelines for webmasters. Content types accepted, including conference papers; PDF searchable-text, crawl-depth and metadata requirements. Read 26 September 2026.
- research Chapekis, A., Bestvater, S., Remy, E. and Rivero, G., 2024. When Online Content Disappears. Pew Research Center, 17 May 2024. Roughly 90,000 pages sampled per year 2013–2023 from Common Crawl; accessibility tested October 2023.
- operator YouTube Help, Use automatic captioning. Speech-recognition captioning, supported languages and the platform's own accuracy and availability caveats. Read 26 September 2026.
- standards body DOI Foundation, What is a DOI?. Definition, persistence and resolution; the statement that persistence rests on organisational commitment. Read 26 September 2026.
- academic Suzgun, M. et al., 2026. Evaluating Commercial AI Chatbots as News Intermediaries. arXiv preprint 2605.22785; 2,100 questions put to six chatbots, 9–22 February 2026. Retrieval failures account for over 70% of errors; Anglophone retrieval bias. Preprint, not peer-reviewed; not a study of conferences.
- industry initiative Think. Check. Attend., introduced by Knowledge E. Three-step framework and conference-checking tool for assessing conference legitimacy. Knowledge E sells services to scholarly publishers, so it is an interested party; the full checklist could not be retrieved. Read 26 September 2026.