AI citations have a half-life. Profound's new Citation Decay feature shows that half of the content cited by answer engines is less than 13 weeks old, which makes content refresh a core part of GEO strategy. Paris Childress also covers Ramp spend data on GEO software buyers, a Machine Relations analysis of citation leaders that miss entire engines, MarketScale's B2B mention and citation split, Leoprd data on how rarely brands are AI's first choice, a local AI recommendations test from Andrew Ryan Marketing, Google's Flipkart buy button test in India, Gemini Skills replacing Gems, and Ipsos entering AI visibility.
AI citations have a half-life. Profound's new Citation Decay feature shows that half of the content cited by answer engines is less than 13 weeks old. In classic SEO, rankings tended to stick. In GEO, a page's likelihood of staying cited declines within a few months, so it needs to be refreshed or replaced.
In this solo roundup, Paris Childress, founder of Hop AI and co-founder of GEOforge, covers nine stories. The theme: the GEO moat is moving from "can you track AI" to "can you prove what changed and why."
Not long, according to Profound. Citation Decay, launched inside Answer Engine Insights, calculates daily citation share on a rolling two-week average and tracks a URL's first citation, rise, peak, and half-life, filterable by model and by owned or all domains. Profound says half of the content cited by answer engines is less than 13 weeks old. The question is shifting from "did we get cited" to "how long did we stay cited, and when should we intervene."
Ramp's September category data puts Profound at 52% purchase incidence among Ramp customers buying GEO software, followed by Peec AI at 18%, AirOps at 17%, Scrunch at 9%, and AthenaHQ at 4%. Profound is up 16 percentage points year over year, and AirOps is the most switched-to vendor with a 29% competitor switch rate. The data covers anonymized spend across 70,000 businesses; the percentages are not total market share, and buyers can purchase multiple tools.
Machine Relations analyzed 96 category question leaderboards across six AI engines. Only 49 winning domains were cited by all six, 47 of 96 were absent from at least one engine, and three still ranked number one in their segment despite being absent from Gemini, Google AI Mode, and Google AI Overviews. A single cross-engine leaderboard can hide complete failure on an important discovery surface.
No. MarketScale's September 28 snapshot covers 420,822 responses, 11,030 prompts, 152 projects, and five engines. In its tracked B2B sample, ChatGPT mentioned brands in 40% of responses but recorded a 0% brand-controlled citation rate. Gemini measured 25% mentions versus 12% brand citations, and Perplexity 22% versus 25%. MarketScale's citation measure means a link to brand content, and the data describes its sample, not all B2B brands or AI answers.
Leoprd analyzed 31,200 AI responses across 27 brands, seven categories, and eight platforms. On brand-agnostic recommendation questions, brands were missing in 36% of answers, mentioned but passed over in 35%, and the first choice in only 29%. Only 40% of repeated tests returned the same recommendation every time, a reminder to run prompts several times, and 86% of cited sources were external to the brand's own channels.
Not reliably. Andrew Ryan Marketing sent the same local buyer query to four AI engines five times across 12 business categories in three US cities, producing 720 answers. Of 673 resolvable recommended businesses, 60.5% appeared in neither Google's local 3-pack nor its organic top 10, and ChatGPT, Perplexity, and AI Mode agreed on the same number one business in only 6% of comparable runs. The test ran on one day in three cities.
The thread through the episode: tracking AI visibility is no longer enough. Brands need to know how long citations last, which engines they are missing, and whether being mentioned turns into being cited.
Subscribe to The GEO Show wherever you get your podcasts, and watch the full episode on YouTube: https://www.youtube.com/watch?v=txnL3Yuqowo
Paris Childress: Hey everybody, and welcome back to another episode of "The GEO Show," brought to you by GEOforge, full self-driving for AI visibility. And we got some great stories today on September 29th, and we are on episode 50, a big milestone for us. And the big theme for today is that the moat of GEO, it's moving from can you track AI to can you prove what changed and why? And that brings us to our first story, which is coming from Profound.
Paris Childress: Profound launched Citation Decay today inside Answer Engine Insights. It calculates daily citation share using a rolling two-week average, and it tracks a URL's first citation, rise, peak, and half-life. Profound says half of content cited by answer engines is less than 13 weeks old. That's an amazing stat. The feature can be filtered by model and owned all domains, and it's available to all of Profound's customers from today.
Paris Childress: So this conversation is moving from did we get cited to how long did we stay cited and when should we intervene? That's getting much more sophisticated. And the stat that jumps out at me most here is that half of these citations are less than 13 weeks old.
Paris Childress: So that suggests that unlike rankings in SEO, where you got rankings and they tended to stick and you didn't really have to refresh content all that often, this suggests that content refresh is a major part of GEO strategy because there is a half-life to citations, and once it gets even a few months old, its likelihood to stick as a citation will decline, and either that piece of content needs to be replaced by something that's more up to date or it needs to be refreshed. So that is a major finding of this last announcement here, and Profound is doing great here by leaning into this with citation decay.
Paris Childress: Next up, we have some fresh market share data from Ramp. Ramp spend data shows Profound with a wide procurement lead. Ramp September AEO category data puts Profound at fifty two percent purchase incidents among Ramp customers purchasing AEO or GEO software versus Peec AI far behind at eighteen percent, neck and neck with AirOps at seventeen percent, and then there's Scrunch at nine percent, AthenaHQ at four percent. Profound is up sixteen percentage points year over year.
Paris Childress: Peec AI is up seven, and Ramp also identifies AirOps as the category's most switched to vendor with a twenty nine percent competitor switch rate. Ramp's dataset covers anonymized spend across seventy thousand businesses, and these percentages are not total market share. And of course, buyers can purchase multiple tools. But this is a sign of just how dynamic this category is right now. Extremely dynamic. Lots of people switching to AirOps. Profound seems to be even building on its dominant lead in this category.
Paris Childress: Peec AI and AirOps are neck and neck battling for that number two spot. And this is a large sample. Seventy thousand US businesses is more than just looking at enterprise spend, but that's a pretty good cross section. So I really think that this market is gonna continue to evolve.
Paris Childress: It's gonna continue to stay really, really dynamic, and I think that there's gonna be a lot of space for a lot of players here as people are still really trying to experiment with this brand new channel and willing to test lots of different tools.
Paris Childress: Next story. Half of citation leaders still fail to reach every engine. Machine Relations analyzed ninety six category question leaderboards across six AI engines. Only forty nine winning domains were cited by all six. Forty seven of ninety six were completely absent from at least one engine, and three still ranked number one in their segment despite being absent from Gemini, Google AI mode, and Google AI Overviews simultaneously. So a single cross engine leaderboard can hide complete failure on an important discovery surface.
Paris Childress: And this is another lesson here that it is important to measure visibility by engine and not look at everything on aggregate because the results are very, very different and it is very hard actually to be present on all of the AI engines.
Paris Childress: Okay, next story. Four hundred and twenty thousand eight hundred and twenty two B2B responses reinforce the mention citation split. MarketScale's September 28th snapshot now covers four hundred and twenty thousand eight hundred and twenty two responses, eleven thousand and thirty prompts, one hundred and fifty two projects, and five engines. In its tracked B2B sample, ChatGPT mentioned brands in forty percent of responses but recorded a zero percent brand controlled citation rate.
Paris Childress: Amazing, forty percent of brands showing up in the response, but zero percent of those brands in the citation of those responses. Gemini measured twenty five percent mentions versus twelve percent brand citations, and Perplexity measured twenty two percent versus twenty five percent. And MarketScale explicitly says that the data set describes its sample, but not all B2B brands or AI answers. So the mention frequency and evidence ownership can behave very, very differently by engine.
Paris Childress: And importantly here, MarketScale's citation measure means that it is a link to brand content. It's not the absence of third party citations, so that's important. So what we're seeing here is a strong divergence, non correlated effect between your brand being mentioned and being cited at the same time. Turns out they're very, very different things.
Paris Childress: Next story, which again demonstrates how difficult this channel is to measure accurately. Brands are AI's first choice in only twenty nine percent of recommendation answers. Leoprd, spelled L-E-O-P-R-D, analyzed thirty one thousand two hundred AI responses across twenty seven brands, seven categories, and eight platforms, plus four hundred and twenty thousand nine oh three anonymized consumer products. On brand agnostic recommendation questions, brands were missing in thirty six percent of answers, mentioned but passed over in thirty five percent, and the first choice in only twenty nine percent.
Paris Childress: Only forty percent of repeated tests return the same recommendation every time, and eighty six percent of cited sources were external to the brand's own channels. So we have several interesting nuggets here. Let's pick them apart one by one. When users are making brand agnostic recommendation questions, prompting LLMs for recommendations, only about a third of the time are, or actually brands were missing about a third of the time, or thirty six percent.
Paris Childress: Mentioned but passed over, I don't really know what that means, about another third, and the remaining third, they were their first choice, which might mean that they are prominently displayed here. But there's no guarantee that there will be mentions of brands at all, even if the prompt itself seems to be looking for recommendations, specific recommendations. And the second interesting nugget is that only forty percent of repeated tests return the same recommendation every time.
Paris Childress: And I don't know how many times these were repeated, but that is a good practice to do several repetitions and runs due to the fact that this is probabilistic sampling. We've said that many times before. And eighty six percent of the cited sources were external to the brand's own channels, and that is a signal of how important it is to build citations and work on brand mentions in external third party sites.
Paris Childress: All right, this next story is proving a divergence of local AI recommendations and classic local SEO. Andrew Ryan Marketing sent the same local buyer query to four AI engines five times across twelve business categories in three US cities, producing a total of seven hundred and twenty answers. Of six hundred and seventy three resolvable recommended businesses, sixty point five percent appeared in neither Google's local three pack nor its organic top ten.
Paris Childress: ChatGPT, Perplexity, and AI Mode agreed on the same number one business in only six percent of comparable runs. The experiment was collected on one day and covers only three cities, so the figures should not be generalized universally. But this is a pretty stark example here of just like in GEO, where we see that there isn't really any correlation between ranking in the top ten in classic SEO and having a higher citation rate. We're seeing the same thing with local SEO. AI recommendation and conventional search rankings, while they are clearly connected ecosystems, the ranking is not a reliable proxy for being recommended.
Paris Childress: The next story is about a very interesting test from Google happening in India. TechCrunch observed a limited Google experiment in India where some users see a buy button on selected Flipkart products surfaced inside Gemini in AI mode. The button opens a Flipkart branded checkout flow without leaving the AI interface. The test currently covers a subset of users and products, and Google confirmed that it routinely tests new experiences, but it did not announce a general rollout. This is a fascinating test.
Paris Childress: I do believe that this is a vision for how people will buy in the future, which is that they will start with AI search. They'll get a recommendation. They may have a conversation when they're ready to buy. Instead of clicking over to the merchant's site, they will click and initiate a purchase inside of the AI interface without having to leave. I think that's a smoother flow. It's a better overall user experience for the buyer. And what does this mean for websites?
Paris Childress: Do they simply just become a product database serving a feed to the AI engines? Maybe, maybe so. But if this results in even more sales, in this case more sales for Flipkart, then I'd say Flipkart should think about going headless here and just allowing AI and AI agents to do all the buying anywhere where the buyers prefer to make those purchases and not insisting on all the purchases happening within its own branded web store.
Paris Childress: Next story, Google is replacing Gems with reusable Gemini Skills. TechCrunch reports an in app Google notice saying that Gemini Gems will migrate to Skills beginning on November 17th, with existing Gems converting automatically. Google's current Gemini help surfaces already include skill creation and management. So I used to love building Gemini Gems, just like I loved building custom GPTs in the early days of that wave in ChatGPT and Gemini. But I have to say, I don't really use them anymore.
Paris Childress: I definitely still use my Claude skills on a daily basis. So this is a smart move by Google to sunset the concept of Gems and just transition that over to Gemini Skills. And I think that here AI experiences are becoming more modular, persistent, personalized, and agentic. So stateless monitoring of one clean prompt increasingly fails to reproduce what an actual user's agent may do.
Paris Childress: Next up, Ipsos enters AI visibility with consumer insight distribution. Ipsos is introducing Ipsos Synthesio AI Visibility today, built with Mention Lab. The solution combines consumer questions across the purchase journey with visibility monitoring across ChatGPT, Gemini, Claude, and other models, and includes competitor comparisons, source influence, and generative engine optimization opportunities. So now we've seen GEO expanding into areas like PR, and now it's expanding into market research, and it's going after consumer insight budgets, not merely SEO, content marketing, and communications.
Paris Childress: So it's another sign that GEO is starting to effectively encroach and eat lots of these adjacent marketing disciplines, market research, consumer insights, PR. I think it is definitely encroaching into brand as well. So that's why I believe that GEO is on its way to being massively bigger than SEO, which has always been a fairly narrow but very important performance marketing discipline. All right. That's it for today. Episode 50 is in the books. Thank you all for tuning in, and we'll see you all on the next one.
