Profound raised $180M at a $1.8B valuation in a Series D co-led by Sequoia and Kleiner Perkins, nearly doubling from roughly $1B in under seven months after its Series C, with TechCrunch reporting revenue tripled in six months and more than 1,000 enterprise customers. On episode 34 of The GEO Show, Paris Childress covers that funding surge plus BrightEdge CDN-edge markdown for AI agents, Google's concession that AI results do not fit legacy position metrics, third-party listicles, Brandy's human vs AI citation gap, directories on supplier prompts, sampling rigor, and why mentions and citations diverge.
Profound's $180M Series D at a $1.8B valuation is a rising-tide moment for generative engine optimization. On episode 34 of The GEO Show, Paris Childress opens with Sequoia and Kleiner Perkins co-leading a round that nearly doubles Profound from a roughly $1B unicorn valuation in under seven months, with TechCrunch reporting revenue tripled in six months and more than 1,000 enterprise customers. The rest of the briefing turns that capital signal into operating guidance: CDN-edge markdown for AI agents, why Google Search Console will not give you AI "positions," why third-party listicles and sector directories still dominate cold-start citation paths, and why mention rate is not citation rate.
Around that throughline Paris covers BrightEdge Agent Edge on Amazon CloudFront, John Mueller on AI blocks versus ranked results, a Search Engine Land cold-start listicle experiment, Brandy's human versus AI citation gap, citation.press directory findings, Benchmark Bureau sampling stability for CRM recommendations, and MarketScale's 399k-response B2B index on mention/citation divergence.
Because capital and enterprise adoption are validating the category, not only one vendor. Profound raised $180M at $1.8B, co-led by Sequoia and Kleiner Perkins, less than seven months after a $96M Series C near a $1B valuation. Revenue tripled in six months per TechCrunch. Customer count is past 1,000 enterprises. Paris notes the public pricing page is gone and reads that as an upmarket push toward a broader marketing platform of record for brand presence in AI. His framing: a rising tide that lifts many boats in GEO, including smaller ones.
This is great for the category. I think it basically is the rising tide that lifts all boats in this category.
Most when the site is JavaScript-heavy. BrightEdge Agent Edge for Amazon CloudFront detects 650+ AI agents and can serve markdown to those agents while humans and traditional crawlers still receive HTML. A vendor-reported case cites a 30% lift in direct AI referral traffic. Cloudflare offers a parallel double-serve path. Paris has not seen strong evidence that AI bots struggle with clean HTML, but they do struggle with JavaScript. Pure HTML sites may see little crawl benefit from the conversion alone.
Because Google treats an AI feature as a block. John Mueller acknowledged that position is difficult to make useful for AI results. Search Console's Generative AI report focuses on impressions (effectively citations), pages, countries, devices, and dates, not ranked positions inside the block. You are in it or you are not. Paris's practical tip: Bing Webmaster Tools already reports query fanouts; re-optimize cited pages against those fanout queries.
Yes, in the studies Paris covers on this episode. Search Engine Land reported 85.8% of 437 source mentions from third-party listicles versus 14% from a brand listicle and 0.2% from PR, with roughly half of cited sources disappearing within 30 days. citation.press analyzed 3,850 commercial prompts and found sector directories supplied 41% of citations versus 18% brand-owned, with structured directories cited 3.1× more than free-text listings (commercial-interest caveat applies). Paris's operating answer: chase third-party listicles with swap agreements, and treat category directories as machine-readable trust surfaces again.
Yes. MarketScale's updated B2B marketing index covers 399,088 AI responses, 112,677 brand mentions, and 79,474 citations. ChatGPT named brands in roughly 30% of responses but cited sources in roughly 20%. Gemini showed about 25% brand presence versus an 11% citation rate. Overall, 68% of observed brand mentions were citation-backed. The relationship varies by engine. Paris argues against collapsing mention and citation into one blended share of voice. GEOforge's overall SOV currently weights mentions 3:1 versus citations as a temporary compromise, while still preferring engine-level split reporting. Related rigor note: Benchmark Bureau's multi-wave CRM sampling (HubSpot 86%, Salesforce ~60%, Zoho 53%; Kendall tau-b 0.861-0.940) and SignalForge's 40-run protocol show probabilistic recommendations can be measured defensibly when sampling is serious.
00:20. Profound Series D: $180M at $1.8B, Sequoia and Kleiner Perkins co-led.
02:25. BrightEdge Agent Edge: 650+ AI agents, markdown vs HTML at the CDN edge.
06:07. Third-party listicles: 85.8% of 437 cold-start source mentions.
07:40. Brandy: top human pages average 682 citations vs 119 for top AI pages.
13:45. MarketScale: 399,088 responses show mentions and citations diverging by engine.
The episode closes on the same operating rule that runs through the funding headline: treat GEO as a category being capitalized, measure with rigorous sampling, and keep mention and citation signals separate.
Paris: Hi, everybody. Welcome back to another episode of The GEO Show, brought to you by GEOforge, full self-driving for AI visibility. We've got some great stories to cover today on September the 16th, episode thirty-four, and it starts with a major funding announcement from Profound. Profound has raised a hundred and eighty million dollars at a one point eight billion dollar valuation.
Paris: Profound closed this Series D, a hundred and eighty million raised at a one point eight billion valuation, co-led by Sequoia and Kleiner Perkins, less than seven months after its ninety-six million dollar Series C, which I believe valued them at about one billion. So almost doubling their valuation in seven months.
Paris: Profound says that it now serves more than one thousand enterprise customers, and TechCrunch reports that revenue has tripled in six months. This is a major, major development for this nascent category of GEO, and it may be even bigger. I mean, the category is expanding because we have Profound leading the way and literally blazing the trail for this category.
Paris: First becoming a unicorn about six or seven months ago and now having that valuation doubled in just seven months. I checked Profound's pricing page. They are no longer displaying any pricing, which is interesting. I think they're trying to move upmarket into enterprise. They are absolutely trying to become the marketing platform of record, not just a GEO or AI visibility management platform, but a full suite of managing your brand's entire presence with regard to AI.
Paris: So big news. This is great for the category. I think it basically is the rising tide that lifts all boats in this category. And believe me, there are many, many boats. We're one of them, and we're one of the small ones. On to the next story. Another big player, BrightEdge, moves AI optimization to the CDN edge.
Paris: What does this mean? BrightEdge's Agent Edge for Amazon CloudFront detects six hundred and fifty plus AI agents and can serve AI agents a lightweight markdown representation while humans and traditional search crawlers continue receiving the normal HTML page. Its AWS Marketplace listing describes crawler analytics, mentions, citations, and AI visibility measurement.
Paris: BrightEdge cites a customer with a vendor-reported thirty percent increase in direct AI referral traffic So this is moving, this is moving GEO really into the content management system infrastructure and content delivery networks. What they're saying here is that they can convert HTML to Markdown, which is a f- more friendly format for AI agents to crawl and ingest.
Paris: And maybe that means they can do it faster, which might mean within a limited crawl budget, they might be able to crawl more pages. I haven't really seen any evidence yet that AI bots struggle to crawl and understand HTML, and I don't really see a need to do this. I looked at this feature with Cloudflare 'cause Cloudflare also has the same feature.
Paris: It's like double serving, if anybody remembers that in the early SEO days, where you serve bots and crawlers one version of the page, and you serve human users another version of the page. One thing that is well-established so far is that AI bots do struggle to crawl and understand JavaScript. So if you do have a JavaScript-heavy website, then probably this solution, converting that into Markdown representation, is a good idea.
Paris: But if your site is predominantly HTML-based, I don't think this is really gonna move the needle and help you get more crawling from AI bots. That's just my opinion. Let's move to the next story. Google concedes AI visibility does not fit legacy position metrics. None other than Google's John Mueller, who is Google's official voice for the SEO world, acknowledged that position is difficult to make useful for AI results because Search Console tracks an AI feature as a block rather than reporting an individual source's rank inside it.
Paris: Google's dedicated Generative AI report currently focuses primarily on impressions, which are effectively citations, pages, countries, devices, and dates, but we don't have any data about ranking positions. I don't think we ever will because, as John Mueller has said, this is a block. They're looking at an AI overview as a block, and either you're in it or you're not in it.
Paris: So if you're in it, you have an impression or a citation for that page, and that is what's currently reported. Or you're not in it, and that page gets zero impressions. Uh, the countries and devices are kind of interesting. What would be very interesting is if Google Search Console followed what Bing Webmaster Tools is doing and actually started reporting on the query fanouts.
Paris: Because then what you can do is you can re-optimize those existing pages with the queries that are reported in the query fanouts, and that would further improve the performance of those pages. That's a great little tip there that you can do with Bing Webmaster Tools. Next story, third-party listicles dominate a cold start GEO experiment.
Paris: All right, we're back to listicles. As reported by Search Engine Land, a 30-day experiment reported 85.8% of 437 source mentions from third-party listicles versus just 14% from the brand's own listicle and 0.2% from PR. Ouch. The broader two-experiment write-up also reported substantial citation turnover, with roughly half of previously cited sources disappearing within 30 days.
Paris: All right, a lot going on here. First, we have dominated statistics from listicles, and not from a brand's own listicle, but from third-party listicles. So whatever people are talking about listicles still are. I think unfortunately a must-have in your citation building strategy. What we do is we have our own listicle, and we're not necessarily trying to push a lot of citations for that listicle, but we use that listicle page as a bartering chip when we reach out to other sites who have published listicles that we wanna be in.
Paris: We say effectively, "Let's do a swap. We would love to be in your listicle, and this is why we deserve to be in your listicle. And by the way, here's our listicle, and we'll put you in ours as well." And so far, we've gotten a handful of swap agreements like that. So long live the listicle. And on we go.
Paris: Brandy finds a five point seven X citation gap between top human and AI-written pages. Again, bucking the trend, Brandy analyzed approximately nine hundred thousand AI answers and four thousand two hundred and ten cited pages. Its twenty-five most cited human-written pages averaged six hundred and eighty-two citations each versus one hundred and nineteen for the twenty-five most cited AI-written pages, which is a five X difference.
Paris: Authorship was classified using Brandy's proprietary methodology, so this does not establish that human authorship caused the citation difference. But we do have an interesting correlation here. And I think that I wanna remind everyone that that purely scaled AI-generated content alone is a losing strategy.
Paris: This is very unlikely to produce the strongest citation assets. That said purely human-written pages right now I think are just no longer practical. You, you have to use AI in some way to move faster, to publish more, not necessarily to scale the hell out of it, but to move faster. So I think that pretty much all of us are operating in this gray area. Somewhere between purely AI-generated content, which is a bad idea, and purely human-generated content, which basically is, is a major competitive disadvantage in today's world.
Paris: So this Brandy study is showing that the human written content did perform better. It got many more citations. So I think the, the best middle ground here is to try to achieve the right balance of scale and human, not human-generated, but human-edited content, and that's the approach that we take.
Paris: So we always are emphasizing in our content proprietary evidence, expert inputs, original datasets, and information gain. And if we do that, we still feel that creating that content which is heavily assisted by AI, as long as it contains that information gain, we still think we have the best chance at citations.
Paris: Next story. "Directories supplied forty-one percent of citations on supplier selection prompts." All right. Directories are back, it appears. citation.press analyzed three thousand eight hundred and fifty commercial prompts across ChatGPT, Google AI mode, Perplexity, and Google AI Overviews. It reports sector directories supplied forty-one percent of citations versus eighteen percent for brand-owned pages.
Paris: Structured directories were cited three point one times more often than free text listings. Citation.press has a commercial interest in citation infrastructure, so the findings should be independently replicated. Okay. Directories, which used to be a very popular link-building tactic to get listed in directories, apparently these are back specifically with category-specific directories.
Paris: It appears they can now function as highly influential machine-readable trust services. So if you take away anything from this episode with regard to where you should focus your off-site citation building efforts, we know listicles are still a thing. Go after those listicles and directories as well. All right.
Paris: The last story we have for today is this one Repeated sampling produces meaningfully stable recommendation benchmarks. I'm sorry, this is not our last story. It's our second to last story, and I'm checking that I am on the right slide. Okay. Once again, repeated sampling produces meaningfully stable benchmark recommendations.
Paris: Benchmark Bureau ran six hundred accepted answers across two hundred US B2B CRM prompts in three timed waves. HubSpot was recommended in eighty-six percent of responses, Salesforce in close to sixty percent, and Zoho in fifty-three percent. So apparently this was... these were prompts in the CRM space.
Paris: Recommendation rank stability between waves showed Kendall tau b values of zero point eight six one to zero point nine four zero. I don't know what that means, but I think it has to do with statistical relevance. And the benchmark used one OpenAI responses API configuration. So this does not represent cross-engine consensus.
Paris: This is for OpenAI only So probabilistic AI recommendations can still be measured defensively when the sampling methodology is rigorous, and that is the key here. The sampling methodology has to be rigorous, and that's what they did here. They ran many, many runs so that they got the sampling error down to a small margin of error.
Paris: And that is exactly the approach that we take with SignalForge when we measure share of voice in our SignalForge module. We actually do forty runs every time we measure a prompt's visibility. We don't run it once the way most other tools do. We run it forty times because we found that it takes that many runs to basically smooth out the volatility and variability curve and to get to a margin of error below plus or minus two percent and a confidence level of above ninety-five percent.
Paris: So Benchmark Bureau is doing it the right way with rigorous sampling. Good job and kudos to them. All right, the last story of today. A three hundred and ninety-nine thousand response B2B benchmark study shows that mentions and citations diverge, something we've covered exhaustively, and here is another study that proves it.
Paris: MarketScale's updated B2B marketing index covers three hundred and ninety-nine thousand and eighty-eight AI responses. They could have just gone a little bit more to get to four hundred thousand. Also one hundred and twelve thousand six hundred and seventy-seven brand mentions are included in those responses and seventy-nine thousand four hundred and seventy-four citations in those four hundred K responses.
Paris: Across five engines, ChatGPT named brands in roughly thirty percent of responses, but but cited sources in roughly twenty percent of the responses. Gemini showed a wider twenty-five percent of brands in the responses versus an eleven percent citation rate. And overall, sixty-eight percent of observed brand mentions were citation-backed So let's unpack this and dig a little bit deeper.
Paris: ChatGPT named brands in thirty percent of the responses, but they cited sources in only twenty percent of the responses. So this is saying basically that there is not a high overlap or correlation between getting mentioned and getting cited. And we don't know what these responses were. We don't know what category, we don't know what the prompts were.
Paris: But what we see is that brands were mentioned thirty percent of the time, but citation sources appeared only twenty percent of the time. So I'm guessing that these were predominantly top-of-funnel prompts, first of all. But the point of this study is that there is no demonstrable correlation between mention rate and citation rate.
Paris: They are quite different signals, and the relationship varies materially also by AI engine. So it's never a good idea to collapse your mention rate, your citation rate into one blended share of voice. It's always a better idea to split those out and look at them separately. The way that we handle this is we do have an overall share of voice by engine that gives a three-to-one differential, a three-to-one weighting of mentions versus citations.
Paris: So we count mentions as three times more valuable than citations in our share of voice score. That's just a decision that we made. It is also subject to change. But the better approach, of course, is to dig deeper and split those apart. What is my mention rate for this engine? What is my citation rate for this engine, and so forth. All right. That's all we've got for today. Thank you all for listening. See you on the next one.
