Funding and press coverage do not buy source authority in AI answers. A Machine Relations analysis found 17 of 24 newly funded AI companies had no cited answer run across nearly 17,000 eligible runs, even though those 17 had announced $1.25 billion in funding. Paris Childress also covers AISO data linking own-site citations to brand mentions, PR Newswire's GEO brand report, Perplexity's Agent API, Microsoft's Copilot Autopilot, Meta Muse tracking, the open-source ELMO platform, a High Salience study of AI recommendations outside Google's top 10, and SharkNinja's AI agents.
Funding won't buy you AI citations. A Machine Relations analysis of 24 AI companies that announced funding in late September found that 17 of them had no cited answer run at all, even though those 17 had announced $1.25 billion between them. Big funding announcements and the press coverage that comes with them do not instantly produce source authority in AI answers.
In this solo roundup, Paris Childress, founder of Hop AI and co-founder of GEOforge, covers nine stories: PR Newswire bringing GEO measurement into PR distribution, Perplexity retiring Sonar for its Agent API, Microsoft's always-on Copilot Autopilot agent, the Machine Relations funding analysis, AISO data on own-site citations and brand mentions, GEO tools adding Meta Muse tracking, the open-source ELMO platform and fan-out searches, a High Salience study of AI recommendations outside Google's top 10, and SharkNinja's AI agents.
Not on its own. Machine Relations identified 24 self-described AI companies that announced funding between September 21 and September 26 and checked them against nearly 17,000 eligible answer runs and 132,514 citation events. 17 of the 24 had no cited answer run at all. Those 17 had collectively announced $1.25 billion in funding.
The analysis measures whether their domains were selected as sources. It does not mean AI systems do not know or mention these companies.
"So the notion that big funding announcements can boost your AI visibility and your citation presence is not necessarily true." Paris Childress
Strongly, according to AISO. It analyzed almost 35,000 GPT and Gemini monitoring observations across 75 anonymized projects. On unbranded prompts:
Paris notes that other data has contradicted this, but this dataset suggests the best way to get mentioned is to win the citation. Working backward, one of the strongest prerequisites for a citation is getting the page crawled by AI bots. That is the funnel his team reports on for content performance: AI crawl activity, then how often the page is cited, then how that influences the mention rate.
Less and less. High Salience tested 100 applicant tracking software buyer questions in ChatGPT with web search and compared recommended brands with Google's top 10 organic results. Of 388 recommended brand mentions, 333 (86%) were brands whose own domain did not rank in Google's top 10, and only 19% of citations came from domains also ranking in the top 10. The study used one run per question, so answer variance was not measured, a flaw Paris calls out. His read: SEO and GEO are becoming separate disciplines, and ranking on page one of Google will not necessarily move AI visibility.
Retrieval is shifting too. Perplexity retired its Sonar API tiers on September 27 and made its Agent API the primary surface. The agent can compose web search, URL fetching, code execution, MCP connections, finance search, and people search on demand, so the same prompt can trigger different retrieval behavior and rely less on search results.
Because agents are starting to recommend products. Microsoft unveiled Copilot Autopilot on September 25, a persistent cloud-hosted agent with its own identity, memory, computer, workspace, permissions, and governance. SharkNinja CIO Velia Carboni told Diginomica that Salesforce Agentforce agents now handle around 20,000 customer chats per week, including shopping and product selection. Tracking is following: LLMPulse added Meta AI and Muse tracking on September 27, with coverage across 18 AI models and Meta tracking from 12 euros per month.
The thread through the episode: AI visibility is earned through citations, not headlines. Brands that get crawled, get cited, and show up across agents as well as answer engines will be the ones AI names.
Subscribe to The GEO Show wherever you get your podcasts, and watch the full episode on YouTube: https://youtu.be/INB0-iSYbJY
Paris Childress: Hi, everybody. Welcome back to another episode of The GEO Show, brought to you by GEOforge: full self-driving for AI visibility. Today, we've got some great stories on Monday, September 28th, episode 49, so let's get to it.
Paris Childress: The first story is from PR Newswire. PR Newswire brings GEO directly into PR distribution. PR Newswire launched its AEO and GEO brand report in Asia Pacific today. Powered by Cision-owned Trajan and embedded in PR Newswire Amplify, it measures brand mentions, model-specific visibility, sources, URLs, and actual model answers.
Paris Childress: Customers can then move directly from identifying a visibility gap into PR Newswire's multi-channel amplification workflow. I checked out this product, and it is very interesting, PR Newswire's Amplify, and it is proof that GEO is no longer, and really never was, an SEO-only software category. Here we have a major communications infrastructure provider that is combining measurement plus narrative intelligence plus distribution in one workflow. And this is why I think that PR agencies are in a great position to move into the GEO space because of the distribution.
Paris Childress: And we know that off-site citation building is probably the most important and the most impactful thing that can move the needle on your AI visibility, and a big part of that is PR-style outreach, so this was a really interesting move here by PR Newswire. I even requested a demo for their PR Newswire Amplify product, and will report back after I dig a little bit deeper.
Paris Childress: Next up, Perplexity completes the Sonar Agent API cutover. Perplexity retired its Sonar API tiers on September 27th and made Agent API the primary surface. Unlike Sonar's fixed model retrieval setup, Agent API can compose web search, URL fetching, code execution, MCP connections, finance search, and people search as tools on demand. Perplexity says existing Sonar tiers map to Agent API presets, though developers have raised backward compatibility concerns around Sonar Pro behavior.
Paris Childress: So Perplexity is really out in front in innovating how the search and retrieval layer functions. So as most of you probably know, when a user prompts an LLM like ChatGPT or Perplexity, in most cases, they go out and do what's called a real-time search retrieval, and this is heavily reliant on search engines today, but that is changing. So now, what we're seeing more from Perplexity is that a prompt can produce different retrieval behavior because now the agent decides which tools to invoke, and this is true even without any kind of an obvious model name change. So that is gonna make these engines less reliant on search results.
Paris Childress: That means SEO and classic rankings will progressively become less and less influential on these real-time search retrieval behaviors because there will be other channels and other methods like MCP and retrieval from databases that are not necessarily coming from search.
Paris Childress: Next up, Microsoft turns Copilot into an always-on digital coworker. Microsoft unveiled a rebuilt Copilot around home, code, and Autopilot on September 25th. Autopilot is a persistent cloud-hosted agent with its own identity, memory, computer, workspace, permissions, and governance. It can continue work without waiting for another prompt.
Paris Childress: Microsoft is also grounding Copilot more deeply in Microsoft IQ, Dynamics 365, Power Platform, and enterprise data. This seems to be Microsoft's answer to GrokBot, Meta Muse, and this very major shift we're starting to see now towards persistent private agents, away from just public answer engines and into always-on agents that operate with their own computers in the cloud. And it certainly makes a lot of sense for Microsoft because you can plug what they're calling Autopilot, a private AI agent, into Microsoft's entire ecosystem. So for those companies that are on Microsoft's full stack, this really makes a lot of sense, and I think this is gonna be a compelling product.
Paris Childress: So agents are coming. GrokBot is here. Meta Muse is here. Microsoft is getting their horse into the race with Autopilot, and we are waiting for Google, OpenAI, and Anthropic, who are all likely to follow suit quickly with their own agents.
Paris Childress: And this is also gonna really impact GEO because right now GEO is about having your brand appear in an answer that is triggered by a user prompt in an LLM. But more so that kind of conversation is gonna be with agents, and that experience is going to be different.
Paris Childress: Next story. 17 of 24 newly funded AI companies had zero cited answer runs. Machine Relations identified 24 self-described AI companies that announced funding between September 21st and September 26th, and checked them against nearly 17,000 eligible answer runs and one hundred and thirty-two thousand five hundred and fourteen citation events. 17 of those 24 had no cited answer run at all.
Paris Childress: Those 17 companies had collectively announced one point two five billion in funding. The analysis measures their domains being selected as sources. It does not mean AI systems do not know or mention these companies. So the notion that big funding announcements can boost your AI visibility and your citation presence is not necessarily true.
Paris Childress: These funding announcements, press coverage that goes with them doesn't guarantee or instantly produce an answer layer source authority. I think that's pretty interesting.
Paris Childress: Next story. Back to a familiar topic. Own-site citations strongly co-occur with brand mentions. AISO analyzed almost thirty-five thousand GPT and Gemini monitoring observations across seventy-five anonymized projects.
Paris Childress: On unbranded prompts, GPT mentioned the target brand in forty-nine percent of answers that cited its own site versus just two point eight percent when the site was not cited. Gemini showed fifty-eight percent versus three point eight percent, meaning fifty-eight percent mention rate when the brand was cited versus three point eight percent mention rate when the brand was not cited. So what is this telling us? Well, according to this data, and we have seen other data that contradicts this, but this data strongly suggests that the best way to get mentioned, to have AI mention your brand in its answer, is to make sure that you win the citation, that you are listed as one of the sources that AI is pulling from when it delivers that synthesized answer.
Paris Childress: And if we move backwards in that process flow, I think one of the strongest prerequisites for getting a citation is to make sure that your page gets crawled by AI bots. And this is kind of the funnel that we're looking at when we report on content performance. We first look at a page's AI crawl activity, and then we look at how often that page is cited in AI answers, and then finally, how that's influencing the mention rate.
Paris Childress: Next up, meta agent tracking is already becoming a commodity feature in GEO tools. LLMPulse added Meta AI/Muse tracking on September twenty-seventh. It now advertises coverage across 18 AI models and search experiences with meta tracking starting at just 12 euros per month as a self-serve add-on. The system measures Meta AI through Muse Spark plus Meta web search.
Paris Childress: It does not claim to reproduce every personalized long-running Muse agent session. And in addition to that, Peak added Muse coverage last week. LLMPulse followed just days later. So now we're seeing the first cases of GEO monitoring tools now adding Meta Muse to their tracking, and I do wonder if GrokBot is gonna be in there soon, but this is another sign that AI agents are becoming extremely popular.
Paris Childress: People are maybe shifting parts of their workflow, parts of their AI search behavior from classic AI search or classic AI chatbots into these AI assistants and AI agents like Meta Muse. And so this is gonna be important, as we mentioned earlier, to start treating these AI agents as legitimate surface areas in their own right. And we really need to think about the GEO visibility and AI visibility across all of these surface areas, starting with LLMs and then into search engines, which can sometimes show AI overviews versus regular links, and now these agents, starting with Meta Muse, which is by far now the most popular one.
Paris Childress: Next story. Open source AEO makes core measurement auditable and nearly free. ELMO is an MIT-licensed self-hostable AI visibility platform whose public source code exposes its mention detection, citation extraction, aggregation methodology, and data model. It stores raw engine responses, fan-out searches, citations, and model versions in PostgreSQL.
Paris Childress: Self-hosting is free and managed cloud starts at $29 per month. A September 27th marketplace listing marks its public launch there, although the GitHub project itself has an earlier development history. So what we see here is a basic prompt execution, citation extraction, share of voice. Interestingly, fan-out capture, which I think is a really interesting feature that I'm gonna start looking more into, and even methodology transparency, these are all now moving toward open infrastructure.
Paris Childress: So this is putting a lot of downward pressure on pricing in the GEO platform category when you have open source players really driving the cost down almost to free. And what I think is most compelling here is analyzing fan-out searches, because one popular workflow in GEO is after publishing a page, analyzing the fan-out searches that are related to that page's citations, and you can do this with, probably with this tool, ELMO, and with others that are already out on the market, and then go back and re-optimize that piece for those fan-out search queries wherever it makes sense, 'cause that's gonna make them even stronger and more likely to appear when those fan-outs happen again in the future. And one of the places that you can see fan-out search queries is in Bing Webmaster Tools. Google Webmaster Tools or Google Search Console does not have this yet, but Bing Webmaster Tools does show this.
Paris Childress: Our next story, another popular topic. 86% of AI recommendations went to brands outside Google's top 10. High Salience tested 100 applicant tracking software buyer questions in ChatGPT with web search and compared recommended brands with Google's top 10 organic results for the same questions. Of 388 recommended brand mentions, 333, 86% were brands whose own domain did not rank in Google's top 10.
Paris Childress: Only 19% of citations came from domains also ranking in Google's top 10. The study used one run per question, which is, I think, not statistically sound in my opinion, and so naturally, answer variance was not measured here, so that is a flaw of this survey. But the highlight of this story is we're seeing increasing evidence of a bifurcation among SEO ranking and AI visibility. 86% of these brands who are winning, these are brand mentions.
Paris Childress: 86% of those brand mentions were brands that did not rank a page in Google's top 10. So I think more and more, SEO and GEO are really separate disciplines. I think there's still value for focusing on SEO and ranking if that is still a major traffic source and you're able to stabilize that and it's still important. But don't think that getting your pages, your brand's pages ranked on the first page of Google is going to necessarily influence your AI visibility.
Paris Childress: And on to today's last story. AI agents now handle roughly twenty thousand SharkNinja customer chats every week. SharkNinja CIO Velia Carboni told Diginomica that Salesforce Agentforce agents now handle around twenty thousand customer chats per week. The deployment spans shopping and post-purchase experiences that include product selection, QR code setup assistance, replacement parts lookup, and other service interactions.
Paris Childress: SharkNinja rolled its Salesforce Commerce data in Agent Stack globally following a roughly nine-month rebuild. So here, this supports the earlier story today about the splintering and the growth and diversification of AI recommendation surfaces. So here we have an AI recommendation surface coming from Salesforce Agentforce. This brand, SharkNinja, is processing around twenty thousand customer chats per week, and that includes shopping and product selection.
Paris Childress: So these agents are also in a position to recommend specific products. And we also need to consider agents that are everywhere being a potential salespeople for our brands.
Paris Childress: All right. That will wrap it up for today. Thanks for tuning in and hope to see you all in the next one.
