สำเนาบทสนทนา 🔗
WEBVTT
00:00:00.000 --> 00:00:08.720
Hey, welcome back to Found in AI, the podcast where we talk about how brands can get found, understood, and recommended in AI Search.
00:00:08.880 --> 00:00:12.560
I'm Cassie Clark, an AI Search Visibility Consultant.
00:00:12.800 --> 00:00:18.480
Today is Thursday, November 5th, and we've got a few interesting developments happening this week.
00:00:18.719 --> 00:00:25.440
First, we're digging into why the way you structure your content matters when AI engines are trying to retrieve and use it.
00:00:25.600 --> 00:00:34.320
Then we're looking at Reddit's plans to experiment with more logged-in experiences and what that could mean for brands that rely on Reddit conversations for visibility.
00:00:34.719 --> 00:00:37.039
And we'll wrap up with two quick stories.
00:00:37.280 --> 00:00:45.280
Google bringing more home service booking options into AI mode and a new comparison of search activity on ChatGPT and Google.
00:00:45.520 --> 00:00:46.960
Let's get into it.
00:01:03.679 --> 00:01:14.159
Tom Johnson, a technical writer at Google, he published a piece about content chunking and what makes documentation easier for AI agents to retrieve and understand.
00:01:14.480 --> 00:01:24.640
Now, before anyone runs off and starts rewriting their entire website into 300-word sections, that's not entirely what this article is saying.
00:01:25.040 --> 00:01:30.879
But I do think it raises an important point about content structure that marketers need to understand.
00:01:31.200 --> 00:01:33.120
Here is the basic idea.
00:01:33.359 --> 00:01:47.200
When we talk about AI engines accessing content, we sometimes imagine that an agent visits a page, reads the entire thing from top to bottom, and then understands everything in context.
00:01:47.439 --> 00:01:54.799
Sometimes that is essentially what happens, but in other cases, an agent is working through a retrieval system.
00:01:55.280 --> 00:02:06.000
That system has index documents and they're broken into smaller chunks or pieces so that it can find and return the information that's relevant to a particular question.
00:02:06.799 --> 00:02:17.759
The agent might not receive your entire article, it might just receive a handful of pieces from your article, which creates a very practical problem for content teams.
00:02:18.000 --> 00:02:22.080
Mostly, do those pieces make sense on its own?
00:02:22.639 --> 00:02:28.719
Let's say that you have an article explaining how to choose the best, I don't know, CRM.
00:02:29.120 --> 00:02:42.879
You write a section with the heading something like what you need to know, then you spend the next four paragraphs referring to this process, these requirements, and the documentation mentioned above.
00:02:43.520 --> 00:02:49.280
That might be perfectly understandable to a person reading the article from beginning to end.
00:02:49.599 --> 00:02:56.800
But what happens now when a retrieval system pulls out paragraph 3 without the rest of the article?
00:02:57.120 --> 00:03:00.560
Well, suddenly the information isn't nearly as useful.
00:03:00.879 --> 00:03:08.400
The AI system has to reconstruct what you're talking about, and it may not have enough context to do that accurately.
00:03:08.639 --> 00:03:12.479
That's where the content structure becomes really important.
00:03:13.039 --> 00:03:23.840
Johnson discusses layout aware chunking, which can use elements such as headings to help preserve the organization of a document when it gets broken into these smaller pieces.
00:03:24.240 --> 00:03:32.319
He also discusses Google Cloud's agent search where chunk sizes and retrieval behavior depend on how the system is configured.
00:03:32.560 --> 00:03:33.840
Now this is important.
00:03:34.080 --> 00:03:38.639
We are talking about a specific type of retrieval infrastructure here.
00:03:38.879 --> 00:03:50.000
This is not Google announcing a new search ranking factor, and it's not evidence that writing shorter paragraphs automatically gets your brand cited more often in Chat GPT or Gemini.
00:03:50.879 --> 00:03:56.080
And I can't record this episode and push it out to everyone without saying this.
00:03:56.240 --> 00:04:03.759
I know that content chunking has become a bit of a controversial topic in the GEO AEO space.
00:04:04.080 --> 00:04:09.599
Some people swear by it, others insist it's just a made-up optimization tactic.
00:04:09.919 --> 00:04:15.680
But here's the thing chunking is a real process used inside of these retrieval systems.
00:04:15.919 --> 00:04:24.160
So the bigger question is whether deliberately restructuring your content around these chunks actually improves AI search visibility.
00:04:24.480 --> 00:04:29.920
I think we might need a little bit more evidence, but this article is kind of pointing to that.
00:04:30.240 --> 00:04:39.199
Chunking does kind of help explain why content structure matters, especially when information is being retrieved and then passed on to those AI engines.
00:04:39.600 --> 00:04:45.199
So the question for today, what should content marketing teams actually do with this?
00:04:45.439 --> 00:04:48.800
Well, first, make your headings descriptive.
00:04:49.040 --> 00:05:04.959
So if someone is searching for information about how to choose a CRM, a heading like five factors to consider when choosing a CRM provides much more context than something generic like what you need to know.
00:05:05.680 --> 00:05:09.519
Second, make your sections reasonably self-contained.
00:05:10.000 --> 00:05:25.040
That doesn't mean repeating the same definition in every paragraph, but you should give the reader and the AI agent reading your information enough context to understand the information without having to hunt through five other sections.
00:05:26.319 --> 00:05:34.160
Third, pay attention to how information is presented in things like tables, tabs, and other page elements.
00:05:34.639 --> 00:05:46.399
If your most important answer is buried inside an
Load full transcript
enormous table or a tab that a particular retrieval system doesn't process very well, that could create some accessibility problems.
00:05:46.720 --> 00:05:51.759
And finally, don't confuse structure with simply making your content shorter.
00:05:52.079 --> 00:05:56.160
A detailed, comprehensive resource can still be useful.
00:05:56.480 --> 00:06:01.920
The goal is just to organize it so that the individual sections remain understandable.
00:06:02.560 --> 00:06:12.240
If you've listened to the show for a while, you know that I talk about structure as one of the three components of my FSA framework, fractious structure, and authority.
00:06:12.560 --> 00:06:14.160
This is exactly why.
00:06:14.480 --> 00:06:25.279
We're not just creating content for someone to read, we're also creating content that may be retrieved, summarized, compared, and then reused by different AI systems.
00:06:25.600 --> 00:06:33.360
And if those systems can't easily determine what a section is about, we're making their job harder than it needs to be.
00:06:33.680 --> 00:06:36.560
So the takeaway with this one is pretty simple.
00:06:36.879 --> 00:06:44.079
Write content that works as a complete resource, but make sure that individual sections can stand on their own.
00:06:44.720 --> 00:06:54.160
That is useful for the people actually reading your content, but it can also make the content much more useful in retrieval-based AI workflows.
00:06:55.519 --> 00:07:08.319
Now, let's talk about Reddit, because that platform has become such an important part of the AI search visibility conversation that this next story is really worth paying attention to.
00:07:08.639 --> 00:07:19.120
According to reporting from Search Engine Journal, Reddit is planning to test more logged-in experiences in select markets outside of the US during the fourth quarter.
00:07:19.439 --> 00:07:22.399
Now I want to be very clear about what this means.
00:07:22.720 --> 00:07:29.519
Reddit is absolutely not announcing that the entire platform is going behind a login wall tomorrow.
00:07:29.759 --> 00:07:37.040
The company has indicated that shared links and some access through search are expected to remain available.
00:07:37.759 --> 00:07:39.519
This is mostly an experiment.
00:07:39.600 --> 00:07:46.879
It's not a blanket shutdown of public Reddit content, but the direction is interesting.
00:07:47.519 --> 00:07:56.800
Reddit has become an important place for people to do things like discuss products, compare experiences, and then get opinions from other users.
00:07:57.120 --> 00:08:01.360
And those conversations are really valuable for AI search.
00:08:01.680 --> 00:08:14.879
If someone asks an AI engine or an AI assistant which software is best for a particular use case, they might ask what customers dislike about the product or whether the service is actually worth paying for.
00:08:15.199 --> 00:08:21.600
Reddit discussions can provide information that doesn't always appear on a brand's own website.
00:08:22.399 --> 00:08:24.720
You get into the nitty-gritty on Reddit.
00:08:24.879 --> 00:08:40.480
You get the complaints, the comparisons, the workarounds, the first hand experiences that only live in these conversations of real users, which is what makes Reddit Reddit, and that's really helpful.
00:08:41.120 --> 00:08:48.159
But here's the question that I think brands need to think about when thinking about this logged in experiment that they're doing.
00:08:48.960 --> 00:08:59.679
What happens if more of those conversations become harder for people, search engines and AI engines to access without an account?
00:09:00.159 --> 00:09:03.279
Well, we we don't know the full answer yet.
00:09:03.759 --> 00:09:16.960
If I had to guess, I would say that it depends on how Reddit implements these tests, what remains publicly accessible, and how different AI engines are able to access Reddit content.
00:09:17.440 --> 00:09:23.200
This whole thing though is a reminder that third-party visibility comes with dependencies.
00:09:23.440 --> 00:09:34.159
Brands can participate in those Reddit conversations, they can learn from the conversations and their users, they can even become part of the recommendations that people make to one another.
00:09:34.480 --> 00:09:44.399
But they don't control Reddit's access policies, they don't control the platform design, and they definitely don't control distribution decisions.
00:09:44.720 --> 00:09:49.759
And I think this is where some brands might get into trouble with their AI search strategy.
00:09:50.000 --> 00:09:57.840
They might see that Reddit is showing up in AI answers and they immediately decide they need a Reddit strategy, which might be true.
00:09:58.080 --> 00:10:13.360
I'm not saying that it isn't, but if your entire strategy is built around getting mentioned on one single platform, because that platform happens to be influential right at this moment, you're building on something that you don't control.
00:10:13.679 --> 00:10:18.720
That rented land comparison comes to mind when I'm talking about this.
00:10:19.120 --> 00:10:23.519
We have already seen how quickly AI search citation patterns can change.
00:10:23.759 --> 00:10:28.159
So my advice to all this is not to stop paying attention to Reddit.
00:10:28.320 --> 00:10:29.519
Quite the opposite.
00:10:29.840 --> 00:10:34.639
Reddit is still very important for your marketing strategy in general.
00:10:35.039 --> 00:10:38.240
I would pay attention to the conversations happening there.
00:10:38.399 --> 00:10:50.000
I'd use it to help understand what questions customers are actually asking, what they're frustrated by, and then how they compare those options that are available to them at the time of purchase.
00:10:50.240 --> 00:10:53.519
And I'd also use it to participate where it's appropriate.
00:10:53.840 --> 00:10:58.080
Transparently, of course, you don't want to be labeled as a spammer.
00:10:58.320 --> 00:11:00.879
Just contribute
something when it's useful.
00:11:01.120 --> 00:11:08.799
But please, please, for the love of all things pretty, do not make Reddit your entire AI visibility strategy.
00:11:08.960 --> 00:11:21.919
There are other parts, other platforms that really help make your AI visibility strategy more comprehensive, so that when something like this happens, you still have visibility elsewhere.
00:11:22.240 --> 00:11:28.559
Now, before I wrap up this segment on Reddit, there is one other interesting detail from this reporting.
00:11:28.799 --> 00:11:36.720
Reddit's leadership has also pointed to AI overviews in AI mode as contributing factors in declining search traffic.
00:11:37.120 --> 00:11:40.159
I think we need to stop and think about that for a second.
00:11:40.399 --> 00:11:51.759
We're talking about a platform whose content is useful, like very useful for answering questions, while the way that people reach that content is also changing.
00:11:52.159 --> 00:11:56.879
I think that's a pretty good illustration of the broader shift that we're seeing in search.
00:11:57.120 --> 00:12:06.960
People can get information without necessarily visiting the original source, and that makes it even more important for brands to think beyond traffic alone.
00:12:07.759 --> 00:12:11.279
Okay, let's move into our quick hits for today.
00:12:11.519 --> 00:12:18.879
First up, Google is expanding what people can do directly through AI mode when they're looking for home services.
00:12:19.120 --> 00:12:29.919
According to Search Engine Journal, Google is introducing booking and quote request links through partners including Angie, Thumbtack, House Call Pro, and Jobber.
00:12:30.240 --> 00:12:35.200
Service Titan also plans to be part of the broader integration later on.
00:12:35.519 --> 00:12:49.440
Now, availability and functionality vary by partner, so this isn't every home service provider suddenly getting a universal booking button in AI mode, but the bigger development is really worth noting here.
00:12:49.840 --> 00:12:57.200
Google is moving beyond simply helping someone find a service provider and moving toward what they've been saying all year.
00:12:57.440 --> 00:12:59.200
The future is a gentec.
00:12:59.600 --> 00:13:03.440
Google is now really just helping the user take the next step.
00:13:03.759 --> 00:13:06.480
Think about someone searching for a plumber.
00:13:06.720 --> 00:13:20.240
The traditional process might involve searching, comparing websites, finding contact information, maybe getting on Facebook and asking, you know, the local community who has the best, and then finally reaching out to someone for a quote.
00:13:20.559 --> 00:13:29.759
AI mode is starting to bring more of that process into the search experience, and that has big implications for local businesses.
00:13:30.080 --> 00:13:34.720
It's no longer enough to think about whether your business appears in the results.
00:13:35.039 --> 00:13:46.080
You also need to think about whether your
information is accurate, whether you're presented on the platforms customers use, and whether they can actually take action once they find you.
00:13:46.399 --> 00:13:51.600
Discovery and conversion are becoming very, very closely related.
00:13:51.919 --> 00:13:57.600
And I expect we'll see more examples of that as AI-powered search experiences continue to evolve.
00:13:58.080 --> 00:14:00.799
One more quick update about search volume.
00:14:01.039 --> 00:14:11.039
Search Engine Journal reports that OpenAI's Adam Fry estimated Chat GPT's search like query volume at roughly one-fifth of Google's.
00:14:11.519 --> 00:14:17.440
Or put another way, Google search volume is estimated to be about five times larger.
00:14:17.919 --> 00:14:20.639
Now, there are a couple of caveats here.
00:14:20.879 --> 00:14:27.919
This is an estimate from OpenAI, and the precise methodology behind that comparison isn't really clear.
00:14:28.240 --> 00:14:31.759
We're also talking about two different experiences.
00:14:32.320 --> 00:14:38.639
People use ChatGPT for all kinds of things that don't necessarily resemble traditional web searches.
00:14:38.879 --> 00:14:49.600
For example, my dental hygienist told me just yesterday that she uses ChatGPT to give her decorating ideas based on the items she has on the table in front of her.
00:14:49.919 --> 00:14:59.759
So, that said, I wouldn't take this number and turn it into a precise market share calculation, but I do think that it is helpful context.
00:15:00.080 --> 00:15:10.639
For a long time, conversations about AI search have been framed as though we're waiting for some future moment where enough people start using AI assistance for discovery.
00:15:10.960 --> 00:15:17.919
But the reality here is that substantial search-like activity is already happening there.
00:15:18.320 --> 00:15:22.159
But that doesn't mean that Google is suddenly irrelevant and far from it.
00:15:22.720 --> 00:15:34.720
It just means that brands need to understand that customers have more than one place to ask questions, they have more than one place to evaluate options, and they have other ways to help make decisions.
00:15:35.360 --> 00:15:43.039
Now, one more important call-out about all this query volume isn't the same thing as website referral traffic.
00:15:43.200 --> 00:15:45.120
I know that might be like duh.
00:15:45.759 --> 00:15:57.039
A platform can handle a huge number of questions without sending a proportional number of visits to external websites, which brings us back to something that I say on this show all the time.
00:15:57.279 --> 00:16:06.000
If you are measuring AI search visibility exclusively through referral traffic, you are probably missing a big part of the picture.
00:16:06.720 --> 00:16:09.200
So let's bring these stories together.
00:16:09.440 --> 00:16:12.000
On one side, we have content structure.
00:16:12.240 --> 00:16:18.159
If an AI system retrieves only part of your content, does that information still make sense?
00:16:18.559 --> 00:16:22.000
On the other, we have distribution and access.
00:16:22.320 --> 00:16:26.080
Reddit is experimenting with how people interact with its platform.
00:16:26.320 --> 00:16:30.879
Google is creating more ways for customers to take action inside AI mode.
00:16:31.120 --> 00:16:36.000
And Chat GPT is handling a meaningful amount of search-like activity.
00:16:36.559 --> 00:16:40.000
All of these developments point in the same direction.
00:16:40.320 --> 00:16:47.840
Brands need to think about how their information is structured, where it's available, and what happens after someone discovers them.
00:16:48.080 --> 00:16:51.919
And that work can't live entirely inside an SEO checklist.
00:16:52.080 --> 00:16:58.559
It requires coordination across content, marketing, product, and sometimes even customer service.
00:16:59.039 --> 00:17:05.519
This is all one of the reasons I've recently introduced team training for AI Search Visibility through Cassie Clark Marketing.
00:17:05.680 --> 00:17:12.160
I want to help teams understand how these systems are changing and more importantly, what they can do about it in their day-to-day work.
00:17:12.400 --> 00:17:16.559
If that's something that your team needs, you can find more information on my website.
00:17:16.799 --> 00:17:21.920
And if you want to dig deeper into any of today's stories, I've linked the resources in the show notes.
00:17:22.160 --> 00:17:33.279
You can also subscribe to the Visibility Report, which is my newsletter that goes out on Fridays, where I share more AI search news, research, and practical guidance for improving your brand's visibility.
00:17:33.519 --> 00:17:36.400
You can find all of those things in the show notes.
00:17:36.720 --> 00:17:38.240
Okay, that's it for today.
00:17:38.400 --> 00:17:39.599
I'm Cassie Clark.
00:17:39.680 --> 00:17:42.400
Until next time, stay visible.