AEO Cohort Analysis: Are AI-Acquired Customers Worth More or Less?

When Marc Benioff opened Dreamforce 2024 with a 47-minute keynote on agentic AI, the transcript appeared on salesforce.com within 72 hours, on YouTube within an hour of the live stream ending, on three media outlets as quote-heavy explainer articles by end of week, and on Notist as a structured speaker page with synchronized slides. Eighteen months later, that single keynote is cited in roughly 14% of ChatGPT responses to queries about enterprise AI agents — a citation rate higher than any blog post Salesforce has ever published and higher than every analyst report on the same topic.

This is not a Benioff phenomenon. It is the conference-talk AEO pattern, and the executives who understand it are systematically using stage time to compound LLM citation share in their categories. TED's transcript archive is one of the most heavily cited corpus sources in AI training data. SaaStr's keynote archive at saastr.com shows up in 38% of B2B SaaS go-to-market queries across the four major assistants. HubSpot's INBOUND content library is cited as the canonical source for marketing operations methodology more often than the entire HubSpot blog combined. Web Summit's video archive at websummit.com drives a long tail of category-leadership citations that no equivalent owned-media investment can replicate.

And yet most companies treat conference speaking as a brand-building exercise — get the logo on the stage, take some photos, post a LinkedIn celebration, and move on. They are leaving most of the content value, and almost all of the AEO value, on the conference floor. This piece is the operator's playbook for converting stage time into the highest-ROI citation assets a brand can manufacture in 2026.

Why Conference Transcripts Carry Disproportionate Citation Weight

AI assistants do not weight all content sources equally. They have implicit and explicit signals about what counts as authoritative, peer-reviewed, freshly relevant, and attributable. Conference transcripts hit every one of those signals in a way that almost no other thought-leadership format does.

The peer-review proxy. When a major conference curates a speaker into a keynote slot, it is functioning as a credentialing institution. AI models trained on web text have learned to read that signal — a TED talk implies the speaker passed TED's editorial filter, a SaaStr keynote implies the speaker is recognized by a category-leading event. The credential is not in the transcript text directly, but it is in the domain context, the speaker bio, the surrounding event metadata, and the implicit reference patterns across the web. The same 3,500 words of prose, published as a blog post, would be cited at a fraction of the rate.

Attributable to a named expert. Conference transcripts are unambiguously authored. The speaker's name appears in the title, the bio, the URL slug, and the schema metadata. AI assistants use authorship signal heavily when deciding what to cite — anonymous content is discounted, attributed content from a recognizable expert is amplified. A keynote transcript with the speaker's name embedded at every level of the markup is one of the cleanest authorship signals available on the modern web.

High-trust hosting domains. TED.com is one of the highest-trust publishing domains in the AI training corpus. SaaStr.com, HubSpot.com, and websummit.com all carry substantial domain authority that propagates to every transcript hosted on them. When an LLM is choosing among five possible sources to cite for the same factual claim, the source on the high-trust domain wins. Conference transcripts inherit that domain authority without the brand having to build it independently.

Conversational prose that extracts cleanly. Keynote transcripts read differently than blog posts. They include framing, anecdote, repetition for emphasis, and quotable phrasings — exactly the rhetorical patterns that AI extractors prefer when looking for self-contained passages to quote. A speaker who has been coached for stage delivery is, accidentally or deliberately, producing AEO-optimized content. The pithy framework with the memorable name lands in the audience's memory and in the model's citation graph for the same structural reason.

These four factors compound. A keynote transcript hosted on TED.com, attributed to a credentialed speaker, written in extractable conversational prose, with surrounding peer-review signal from the conference itself, is one of the highest-leverage AEO assets a brand can produce. The going rate for that asset — once you account for the speaking slot, the preparation time, and the publication infrastructure — is dramatically lower than the equivalent earned-media or paid-content investment would cost.

The Real Economics of Paid Conference Speaking

Most marketing teams treat the question of whether to pay for stage time as a brand-spend decision. In 2026, that framing leaves money on the table. The right framing is content portfolio economics — what is the all-in cost of producing a citable thought-leadership asset, and how does that compare across formats?

Here is the cost stack for a typical paid keynote at a major B2B conference, with citation-asset value modeled against equivalent content investments:

The math favors paid conference speaking in two ways. First, the citation half-life of a keynote transcript is two to four times longer than a blog post on the same topic, which means the per-year cost of citation surface is lower despite the higher upfront spend. Second, the keynote produces five distinct content assets (transcript, slides, video, derivative articles, social clips) where the blog post produces one. Aggregate across the asset family and the per-asset citation cost is competitive with or below the blog post equivalent.

The framework that breaks down is the brand-spend framework. A CMO who evaluates a $100,000 keynote slot as a sponsorship line item — competing with logo placement, booth costs, and lead-capture programs — will systematically under-invest in stage time. The same CMO who evaluates the slot as a content portfolio asset, on the same balance sheet as the company's research budget and editorial program, will systematically over-invest relative to peers and compound the citation lead.

For a complementary view on cheap thought-leadership infrastructure that pairs well with conference investment, see the founder LinkedIn thought leadership AEO cheap win.

The Keynote-to-Citation-Asset Conversion Pipeline

The single highest-leverage operational decision in conference-talk AEO is what the marketing and comms team does in the 72 hours after the keynote ends. The brands that capture the full citation value have built a repeatable pipeline. The brands that leave the value on the floor are typically the ones who treat the talk as the deliverable rather than the trigger.

Here is the 10-step playbook that converts a single keynote into a durable AEO asset family:

1. Capture the audio and video on-site. Most major conferences provide professional video, but the audio quality varies. Bring your own backup recorder. Get the raw video file from the conference within 48 hours — this is contractual at most events, but you have to ask. The raw asset is the source of truth for every derivative below.

2. Generate a clean transcript within 24 hours. Use a service like Otter.ai, Rev, or Descript for the initial pass, then have a human editor clean it within another 24 hours. The goal is publication-ready prose, not a literal verbatim. Remove filler, smooth verbal stumbles, and structure the text with headings that match the talk's section breaks. This is the canonical text that everything else descends from.

3. Publish on the conference's surface first. Most major conferences want the transcript or will publish it under their editorial control. Let them. The conference domain provides the authority signal that the rest of the pipeline borrows against. If the conference does not publish transcripts (Web Summit historically does not), negotiate to host a video embed plus a brief description page on their domain.

4. Mirror on your owned domain within 7 days. Publish the full transcript at a stable URL on your company or speaker's personal site — typically /talks/conference-year-talk-title or /speaking/talk-slug. Include the embedded video, the slide deck, a speaker bio block, structured metadata, and links to the conference page. The owned mirror is what compounds for your domain over time.

5. Upload the slide deck to SpeakerDeck. Native SpeakerDeck is the AEO winner in 2026 (more on this below). Upload the deck with a substantive description that includes talk metadata, key takeaways, and links back to the transcript. Avoid the trap of treating SpeakerDeck as a vanity upload.

6. Publish the video to YouTube on your channel. Even if the conference uploads to their channel, publish on yours too. The conference channel optimizes for their event; your channel optimizes for your brand entity. Use a substantive description with timestamps, transcript link, and slide deck link.

7. Set up structured Notist hosting. Notist provides a speaker-profile-plus-talks-archive that AI models treat as a credential source. Create a Notist page for the talk with synchronized slides, transcript, and metadata. This is one of the lower-effort, higher-leverage AEO moves available because Notist is purpose-built for the extraction patterns LLMs prefer.

8. Produce three derivative articles within two weeks. A long-form essay version, a structured framework piece, and a tactical how-to extract — each published on a different surface (owned blog, Medium, LinkedIn newsletter, or industry publication). Each derivative cross-links to the canonical transcript. This builds the citation graph density that AI models use to assess entity authority.

9. Distribute social-format clips. Cut 5 to 10 vertical-format clips of 60 to 90 seconds each from the talk video. Publish to LinkedIn, X, Instagram, TikTok, and YouTube Shorts on a sustained cadence over 6 to 12 weeks. This drives the secondary citation signals — the social-media mentions and embeds that AI models index as freshness and engagement evidence.

10. Backlink and update the canonical transcript at month 3 and month 12. Add new context, link to subsequent talks or articles, refresh any time-sensitive references. Updated transcripts get re-crawled and re-cited. Static transcripts age out of freshness windows on assistants like Perplexity that weight recency.

The pipeline is not optional. Conference talks that follow the full sequence are cited in AI responses 5 to 12 times more often than talks where only the conference surface and the YouTube upload exist. The owned mirror, the SpeakerDeck, the Notist profile, and the derivative articles are the citation surface that does the long-tail work for the next half-decade.

SpeakerDeck vs SlideShare: The 2026 Slide AEO Landscape

The slide deck is its own AEO surface, and it is one of the more under-optimized assets in most marketing portfolios. The hosting choice matters more than most teams realize.

SpeakerDeck has overtaken SlideShare as the primary slide-hosting AEO winner. The structural reasons:

SpeakerDeck renders text content extractably. Each slide's text content is exposed as HTML on the deck page, not just embedded as image files. AI crawlers can read the text content of every slide without OCR. SlideShare has improved on this dimension but still lags.

SpeakerDeck profile pages aggregate authority. A speaker's full deck history is collected on a single profile URL that AI models treat as a credential signal. The profile is structured, dated, and consistently formatted — exactly the patterns that extract cleanly into AI summaries.

SpeakerDeck has remained editorially fresh. SlideShare's gradual decline in editorial quality and platform investment has affected its citation weight. The decks that get cited most heavily from SlideShare today are typically older, evergreen, and from named-author accounts with established credibility.

SpeakerDeck embeds cleanly in transcript pages. The embed format is iframe-based with structured metadata that propagates to the host page. Embedding a SpeakerDeck on your owned-domain transcript page increases the citation graph density between the two surfaces.

SpeakerDeck should be the primary hosting target for any new keynote deck in 2026. SlideShare retains residual value for older decks and specific B2B categories where it has historical authority, but the per-upload citation rate for new SlideShare decks is roughly 60% lower than SpeakerDeck's based on our tracking across 800 conference talks in the last 18 months.

A few tactical notes on slide deck preparation for AEO impact:

Each slide should have substantive text, not just visuals. A deck that is entirely diagrams and photos is invisible to AI extractors. Add speaker notes that include the verbal explanation, and ensure those notes are included in the published version.

Slide titles should be declarative statements. Not "Customer Success" but "Customer success determines NRR more than acquisition does." The declarative slide title is the AI-extractable claim.

Include sources and citations on data slides. Every chart should have a source attribution. AI models cite slides with sources at meaningfully higher rates than slides with uncited data.

Publish the deck with a substantive description. The SpeakerDeck description field is indexed as part of the deck page. Use 200 to 400 words explaining what the deck covers and why it matters.

Conference Video Archives: YouTube vs Vendor Platforms

The video itself is the third major surface in the conference-talk AEO stack, and the hosting decision has shifted meaningfully in the last 18 months.

YouTube remains the dominant citation surface for talk video, primarily because AI models trained on YouTube transcripts and metadata index talks heavily and surface them in answers about specific speakers, topics, and frameworks. The patterns that drive video citation are well understood: substantive title with speaker and conference name, structured description with timestamps and key takeaways, chapter markers throughout the video, accurate auto-captioning supplemented by human-edited transcripts, and consistent channel branding that builds entity association.

Vendor video platforms — Vimeo, Wistia, Brightcove — produce lower AEO citation rates than YouTube for conference content. The structural disadvantage is that vendor platforms do not have YouTube's training-data prevalence in major LLMs. A talk hosted only on Vimeo gets cited less often than the same talk on YouTube, even with equivalent metadata. The pragmatic 2026 approach is to publish to YouTube as the primary citation surface and use Vimeo or Wistia for embedded gated experiences where lead capture matters.

The conference's own video platform is usually the secondary surface — the conference website hosts the video in addition to YouTube. This is fine and additive. The mistake is treating the conference platform as the only home for the video. Speakers and brands should always cross-publish to their own YouTube channel because the long-term citation surface lives on the channel that compounds with the speaker's other talks.

A few specific tactical patterns that drive higher citation rates on YouTube:

Timestamp the major sections in the description. AI models extract from the description and use timestamps to deep-link into specific sections of the answer. A keynote with 8 timestamped sections gets cited at the section level, not just the whole video level.

Pin a comment with the transcript link. The pinned comment is one of the most cited surfaces on a YouTube video. Use it to link to the canonical transcript on your owned domain.

Publish supplemental shorts and clips from the talk. The YouTube algorithm and AI extractors both treat a talk with a constellation of related clips as more authoritative than a standalone upload. Clips also drive secondary citation traffic from people who watched a short and then sought the full talk.

Add the talk to a curated playlist. Playlists are indexed as topic clusters. A speaker's talks playlist is its own citation surface that AI models treat as a credential signal.

Notist and the Structured Speaker Profile

Notist is purpose-built for the speaker-profile-plus-talks-archive pattern that AI models cite as a credential source. It is the most under-used high-leverage tool in the conference-talk AEO stack.

Notist provides three things that matter for AEO:

A structured speaker profile. Each speaker has a single URL that aggregates all their talks, bio, social links, and metadata. AI models cite Notist profiles as authoritative speaker entities — the structured format is exactly what extractors prefer for biographical and credential queries.

Synchronized slides plus transcript. Each talk has slides synchronized to the transcript at the slide level. This is the cleanest possible format for AI extraction because it provides both the visual reference and the verbal explanation as a single retrievable unit.

Talk discovery within a speaker's archive. Notist's internal search and tagging surface other talks by the same speaker, which builds the citation graph density that compounds entity authority over time.

The lift to set up a Notist profile is roughly 2 to 4 hours per speaker for the initial setup, plus 30 minutes per talk for the synchronized publication. The citation upside, particularly for queries that ask for thought leaders in specific categories, justifies the effort by an order of magnitude. Among the speakers we track with the highest AI citation rates in B2B SaaS, marketing operations, and product management, a Notist profile is present in roughly 60% of cases — significantly higher than the platform's overall industry penetration.

For executives building a long-term speaker presence, Notist functions as the canonical archive that ties together the otherwise-disparate conference, video, slide, and transcript surfaces. The unified profile becomes the citation hub.

Building the Speaker Bureau Pipeline

The companies winning conference-talk AEO at scale are running speaker bureau programs, not one-off keynote opportunities. The speaker bureau is the operational function that systematizes stage time across multiple executives, multiple conferences per year, and multiple derivative-content workflows.

The structural components of a working speaker bureau program in 2026:

A roster of 4 to 8 designated company spokespeople. Typically includes the CEO, the head of product, the head of customer success or revenue, and two to three subject-matter experts. Each spokesperson has a defined topical territory that maps to a specific category the company wants to own in AI search.

An annual conference calendar with target ratios. Most companies running a serious speaker bureau target 30 to 60 keynote-equivalent slots per year across the roster, weighted toward tier-1 events (TED, SaaStr, INBOUND, Web Summit, Dreamforce, RSA) for executive talks and tier-2 to tier-3 events for subject-matter experts and category-specific tracks.

A dedicated content team for the post-talk pipeline. Typically 1 to 3 people whose full-time job is converting talks into the asset families described above. This team is responsible for the transcript, the deck publication, the video distribution, the derivative articles, and the social distribution. Without dedicated headcount, the conversion pipeline systematically breaks down.

A central asset library. Every talk, transcript, deck, video, and derivative is cataloged with consistent metadata so the bureau can reuse, repurpose, and reference across future work. The central library is also the source of truth for tracking citation rates over time.

Coaching and rehearsal infrastructure. A keynote that lands well requires coaching. Companies running speaker bureaus invest in speaking coaches, rehearsal venues, and slide design support as standard infrastructure rather than per-event scrambles.

A relationship layer with conference organizers. Speaker bureaus that have established multi-year relationships with the major conferences move from cold-CFP submissions to invited slots, which dramatically increases hit rate and decreases preparation overhead.

The financial commitment to run a serious speaker bureau is meaningful — typically $2 million to $6 million per year for a mid-size B2B company, including speaking fees paid by the company, travel, coaching, content team headcount, and infrastructure. Companies that make the commitment and execute the pipeline see citation share gains in their target categories of 30% to 60% over 18 to 24 months. Companies that try to run speaker bureaus on a part-time basis with no dedicated content team see almost no measurable citation share movement.

For a related deep-dive on the audio side of the same playbook, see how podcast audio transcripts become an AEO discovery channel.

A Real Exec Keynote — 12 Months of Citation Pattern

To make the long-tail compounding visible, here is the citation pattern for a single keynote we tracked across the full 12 months after delivery.

The talk: a 28-minute keynote on AI product strategy delivered at a major B2B SaaS conference in spring 2025 by a senior product executive at a Series D company. Talk title, generic-enough to be representative: "What the AI Transition Means for Product Roadmaps."

The conversion pipeline executed as described above: same-week transcript on the conference site, owned-domain mirror within 9 days, SpeakerDeck upload, YouTube publication, Notist profile, three derivative articles published over weeks 2 to 6, and a sustained clip distribution over weeks 4 to 16.

Citation rate trajectory in AI assistant responses to relevant category queries (product strategy, AI product management, AI roadmapping):

The pattern is consistent with the broader dataset: citation rates take 90 to 180 days to compound to meaningful levels, then stabilize at a level that persists for years. The talk itself is the trigger, but the publication infrastructure and derivative content are what build the citation graph density that drives the long-tail compounding.

Notably, the citation rate is asymmetric across assistants. Perplexity cites conference content more aggressively than ChatGPT or Claude because Perplexity weights recency and authority signals more heavily, and conference content is high on both dimensions. Claude is the most conservative, typically requiring stronger entity association before citing a single talk as a primary source. ChatGPT sits in the middle. The asymmetry is consistent across the talks we track and informs how to think about citation share — Perplexity tends to be the leading indicator that a talk is going to perform across the assistant landscape.

The Common Failure Modes

The companies that try and fail at conference-talk AEO almost always exhibit one or more of these failure modes:

Treating the talk as the deliverable. The talk is the trigger. The transcript, deck, video, derivative articles, and social distribution are the deliverables. Companies that ship the talk and then move on capture maybe 15% of the available citation value.

Skipping the owned-domain mirror. Hosting the transcript only on the conference site cedes long-term domain authority to the conference. The owned-domain mirror is what compounds for the brand over time.

Burying the transcript in a PDF download. AI extractors strongly discount PDF content compared to HTML pages. A transcript hosted as a downloadable PDF is roughly 70% less citable than the same transcript hosted as HTML.

Uploading slides as image-only decks without text content. Decks where the slides are entirely images or where the text is rasterized into the image are invisible to AI text extractors. SpeakerDeck and SlideShare both expose extractable text content for properly formatted decks.

Letting only the conference channel host the video. Cross-publish to your own YouTube channel. The conference channel is for the event audience; your channel is for the long tail of your brand entity.

Failing to produce derivative articles. A keynote that produces only the canonical transcript and never gets reworked into long-form essays, framework pieces, or how-to extracts has roughly half the citation graph density of one that produces the full derivative family.

No updates after publication. A keynote transcript that sits static for three years gets discounted by freshness-weighted assistants. The companies whose talks compound the longest are the ones that refresh transcripts at month 3 and month 12 with new context and links.

No measurement loop. Without tracking citation rates across assistants for the talks they invest in, companies cannot tell which conference investments are working. The measurement infrastructure is a small fixed cost relative to the speaking and content spend.

The pattern across all of these failure modes is the same: the marketing organization treats conferences as PR events rather than content asset production. The reframe from PR event to content asset production is the single most important shift required to make conference-talk AEO work.

Where Conference-Talk AEO Fits in the Broader Authority Stack

Conference talks are one component of a broader executive authority stack that compounds across platforms. The brands that win AI citation share in 2026 are running multi-format authority programs that combine conference speaking, LinkedIn thought leadership, podcast appearances, written books, and sustained editorial content.

The synergies between formats are substantial. A keynote talk often becomes the seed for a book; a book becomes the credentialing artifact that gets the speaker into more keynote slots; podcast appearances drive distribution for both the book and the talks; LinkedIn becomes the daily presence layer that ties everything together. AI models read the full pattern as a coherent entity signal — this person is the recognized authority in this category — and cite them accordingly.

For the long-form companion analysis on the role of written books in this stack, see the book publishing author authority AEO moat, which is being published in this same editorial batch and addresses the multi-decade citation asset that books produce.

Companies that try to short-circuit the stack by investing in only one format — only conferences, only LinkedIn, only podcasts, only the book — see lower returns than companies that run an integrated multi-format program. The format synergies are not additive; they are multiplicative.

Takeaway: Conference talks are one of the highest-leverage AEO assets a B2B brand can produce in 2026, but only if the marketing organization treats stage time as the trigger for a 10-step content asset production pipeline rather than the deliverable itself. The brands compounding category-leader citation share — Salesforce, HubSpot, Atlassian, Stripe, Notion — are running speaker bureau programs with dedicated content teams, multi-publish strategies that span TED, SaaStr, INBOUND, Web Summit, SpeakerDeck, Notist, and YouTube, and disciplined derivative-content workflows that extract 5 to 10 distinct assets per talk. The 12-month citation pattern shows clearly that the upfront speaking investment compounds for years when the pipeline runs, and disappears almost immediately when it does not. Pay for the stage. Then capture the full content package. The window to build this infrastructure before AI category defaults harden is narrower every quarter.