YT YouTube2Text

YouTube2Text → Transcripts → Save Your Business from AI (Do This!)

Save Your Business from AI (Do This!) — Transcript

by Caleb Ulku · 4,219 words · 660 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00A local business can rank in the top
0:01three on Google Maps and still get left
0:03out of Google's AI overview. Being close
0:05to the customer helps you rank on the
0:07map, but that doesn't guarantee an AI
0:10recommendation. Now, Semrush has a video
0:12on getting AI to recommend businesses.
0:14Let's see how this applies if you own a
0:17local business.
0:18>> What if your biggest competitor isn't
0:19another company? What if it's AI itself?
0:21Right now, millions of people are
0:23skipping traditional search and getting
0:25answers directly from ChatGPT, Claude,
0:27Gemini, and Google's AI overviews. And
0:29when AI gives an answer, there's a good
0:31website never gets to click. In fact,
0:33one company felt this shift so
0:35dramatically that it sued Google. That
0:37company was Chegg. Google's AI overviews
0:39were killing its traffic and their
0:41non-subscriber traffic was down 49%
0:43year-over-year. Now, this is an extreme
0:45case, but it reflects something bigger.
0:47AI is changing how people find
0:49information and AI platforms are
0:51becoming a regular part of how your
0:53audience researches products and makes
0:54decisions.
0:55>> Now, Chegg here makes sense because the
0:57information was the product they were
0:59selling. If someone gets the answer from
1:01AI, they don't need to visit Chegg. But,
1:04of course, for a local business, someone
1:06still has to come to the house to
1:07replace the water heater. No matter how
1:09advanced AI is, it's still at least a
1:11couple of years from being able to do
1:12that. But, Google can change which
1:15business is going to get that customer.
1:17In the maps, proximity gives you a huge
1:20advantage. A business with a mediocre
1:22website at best can still rank just
1:25because it's close to the person who's
1:26searching. But, if you move a few miles
1:28away, that same business will disappear.
1:31We are seeing something very different
1:32with the AI overview. The businesses
1:34getting recommended aren't the same ones
1:37that rank in maps. So, you can't assume
1:39being nearby the searcher is going to
1:41carry you into the answer. So, if you
1:44own the business or you're doing a
1:45business's SEO, we need to make sure
1:47that we build enough relevance and trust
1:50that Google has reasons to recommend it
1:52beyond proximity.
1:54>> ChatGPT alone went from 100 million to
1:57over 800 million weekly active users in
1:59about 2 years. That means two things for
2:02marketers. First, Google isn't the only
2:04place people are finding answers
2:05anymore. Second, you need a strategy to
2:07be visible inside AI-generated answers.
2:09In this video, I'm going to walk you
2:11through that strategy. It's called LLM
2:13seeding, and it's how you get your brand
2:15mentioned and cited across AI platforms
2:17like ChatGPT, Gemini, and
2:19>> Do you want to also notice how weird the
2:20logos look for these companies? Like so
2:22kind of
2:23I won't say what they look like.
2:25Yeah.
2:26>> Let's start with the basic. What is LLM
2:28seeding? LLM seeding is the practice of
2:30publishing content in the formats and
2:32places that AI systems are most likely
2:34to find, understand, and reference.
2:36>> So, I actually released a video about
2:38this. Anthropic sort of started this off
2:40by releasing a study that showed only
2:42250 pieces of content on the entire
2:45internet was enough to influence even
2:48frontier models. That study happened
2:51a year or so ago. And like I said, I did
2:53break that down and what the
2:55implications of that study were for
2:56local businesses. But let's see what she
2:58has to say.
2:59>> best accounting software for small
3:00businesses. You'll notice the answer
3:01mentions specific brands, compares
3:04features, and pulls from a range of
3:06sources, like review sites, Reddit
3:08discussions, comparison articles,
3:10YouTube videos. LLM seeding is how you
3:12get your brand into that mix. You
3:14deliberately create content about your
3:15brand and place it across those same
3:18types of sources. Because the more AI
3:20encounters your brand in trusted places,
3:22the more likely it is to mention you
3:24when someone asks a relevant question.
3:25>> But okay, so she's looking at accounting
3:27software. This is something that, you
3:29know, someone is going to buy from
3:31anywhere. But with a local business, we
3:33need to explain the particular service
3:35and why this particular business is a
3:38good choice in this area. And there's a
3:41simple check behind this. We start, look
3:43at the website, take any paragraph from
3:45that website, change the business name
3:47to one of their competitors, and if it
3:49fits on on competitor's website, then we
3:51haven't really said very much about your
3:54business, right? Something like
3:56professional water heater replacement
3:58from experienced technicians.
4:01These are platitudes. Every plumber can
4:03say that. There's nothing unique about
4:04that. Or the fact that every plumber
4:06advertises that they're licensed. I
4:08mean, I should hope that's not a
4:11significant competitive advantage. I
4:12hope every plumber is licensed. The
4:15owner knows what makes an installation
4:17difficult. They know what changes the
4:20price quote, what customers can often
4:22misunderstand, and what problems they
4:24find in the homes, in the neighborhoods
4:27that they actually work on. That's how
4:30we need to write local content. Explain
4:32the service through the experience of
4:34the business. Give someone information
4:36that they couldn't get by reading the
4:38same generic paragraph on 20 other
4:41websites. Very simple rule, when we're
4:43writing local content for a local
4:46business,
4:47we want the content to be written and
4:50feel like it was written by someone who
4:53actually does that service in that area,
4:56okay? I know that sounds like a low bar,
4:59but it's a bar that 80 to 90% of local
5:02businesses that I look at don't meet.
5:05>> How do you actually do LLM seeding?
5:07There's a framework for it, the LLM
5:09seeding framework. AI models tend to
5:11favor three things when deciding what to
5:13site. Content that is well structured
5:15and easy to parse, clear context about
5:18what a brand does and who it's for, and
5:20consistent mentions across multiple
5:22trusted sources. The LLM seeding
5:24framework is designed to build all
5:25three. It works as a three-part cycle.
5:27Part one is publish. You create
5:29citation-worthy content on your own
5:31site. Assets like comparison guides,
5:33detailed reviews, FAQs, and original
5:35research. Part two is distribute. You
5:38take that content and your brand beyond
5:40your own domain, which means third-party
5:41sites, video platforms, review sites,
5:43community spaces. Part three is
5:45reinforce. You keep showing up with
5:47consistent messaging over time because
5:49one-time campaigns won't build the kind
5:50of repeated presence AI needs to feel
5:53confident citing you. Each part feeds
5:54the next and over time the visibility
5:56compounds.
5:57>> publish, distribute, reinforce. The
5:59framework makes sense, but the quality
6:02of the information has to come before
6:04distributing it. If we write this
6:06generic description and publish it in 10
6:08pages, we have 10 copies of a generic
6:11description, okay? So, again, we want to
6:14start with what the business actually
6:17does, what services it offers, where it
6:19works, what customers value about it,
6:22and what makes its approach different.
6:24Then, keep that information consistent
6:26across the business's website and all of
6:29these relevant third-party sites. The
6:30owner explains these things to customers
6:33every single day, but none of that, for
6:36most businesses, ends up being on the
6:38website. That's what we need to get out
6:40of the owner's head and onto their
6:43website.
6:44>> Let's get into the details. Step one,
6:45establish your product entity. The first
6:48thing you need is a clear reference
6:49point for your product on your own site.
6:51One authoritative page that explains
6:53what your product is, who it's for, and
6:55how it fits into your category and other
6:58essential details. This page becomes the
7:00source of truth AI systems can learn
7:02from and cite confidently. I might be
7:04biased here, but a good example of this
7:06is our own AI visibility tool kits
7:08landing page. It starts with a clear
7:10value proposition at the very top, so
7:12both readers and AI immediately
7:14understand what the product offers. It
7:16follows a logical heading structure that
7:18makes it easy to scan and parse. The
7:20benefits are tied to specific use cases,
7:22not generic feature lists, and it
7:24includes FAQ content written in natural
7:26question and answer format because that
7:28mirrors how people actually ask AI for
7:30information. Once you have this page in
7:32place, link it to every relevant piece
7:35of content you publish. Blog posts,
7:37landing pages, press releases, author
7:39bios. This helps AI systems recognize it
7:42as the authoritative source for
7:43information about your product.
7:45>> Yeah, this is very close to how we
7:46structure local business websites. We
7:48call it the core 30 structure. So, let
7:51me talk you through what I mean here.
7:53We'll start with the Google Business
7:54Profile. That's the heart of the entity
7:56for any local business. Uh that GBP is
7:59going to list a variety of categories
8:00and a variety of services. When we're
8:02building up the website, we want to link
8:04all of those service entity pages and
8:06category entity pages together in the
8:09exact same way that they're linked to on
8:11the GBP. So, we're going to create a
8:13page for every category. We're going to
8:15create a page for every service. Then
8:17the secondary categories get linked to
8:19from the home page. The services get
8:20linked to from the parent category
8:22pages. This is the core 30 structure,
8:25very simple, and it mimics the GBP. It
8:28feeds Google and the AI systems that
8:31your website matches your Google
8:32Business Profile. That the entity
8:34relationship that you told Google you
8:37have on your GBP, you're reinforcing
8:40that with your website structure itself.
8:43The structure gives us somewhere to put
8:45all of this information, right? The
8:48Google Business Profile and the website
8:50are describing the same business. If the
8:53profile lists services that the website
8:55never explains, we're missing an
8:57opportunity to build this relevance, to
8:59build this service entity relationship.
9:02Very similar to the structure she gave
9:04us for a software company, because
9:06that's how Google, that's how AIs read
9:09content. That's how they understand
9:11entities. You just get an advantage on
9:13local because you have the ability with
9:15your Google Business Profile to feed
9:18very clear relationships into these
9:20systems.
9:21>> Step two, seed structured narratives
9:23across third-party sites. AI learns from
9:26the wider web and not just your domain.
9:27So, you need content about your product
9:29living on trusted sites in your
9:31industry. There are several ways you can
9:33do this. Work with affiliate partners
9:34who already use your product and ask
9:36them to publish detailed reviews. Pitch
9:38guest posts to high authority industry
9:40publications. Get your product listed on
9:42comparison and review sites cover your
9:44category. The specific tactics will vary
9:46depending on your industry, but at the
9:48bare minimum start with what you can
9:50control. Your website, your landing
9:51pages, your social profile bios, your
9:53descriptions on review sites. Make sure
9:55they all describe your product the same
9:57way. Cover what it does, who it's for,
9:59and what makes it different. That
10:01consistent messaging is what helps AI
10:03build a clear picture of your product
10:05and become more likely to reference it.
10:07>> Of course, she's talking about a
10:08software example and a lot of these
10:09tactics fit that better than they would
10:11fit a local business. If you're running
10:13a plumbing company in a small city, I
10:15would not recommend that you start by
10:17building an affiliate review campaign.
10:19Instead, look at the sources appearing
10:21for the services you want customers for.
10:24If a directory or local publication
10:26keeps getting cited, investigate it.
10:28Make sure that the business information
10:30for your business is on that platform
10:33and is accurate. Look for a reasonable
10:35opportunity to earn relevant coverage.
10:37And the consistent descriptions are very
10:39important. The website, the business
10:41profile, and other listings all need to
10:43agree about what the business does, the
10:45business name, the business address, and
10:47the business phone number. Now, you are
10:49of course working with a much more
10:50limited budget than a large software
10:52company. So, we want to put that budget
10:54into sources that matter instead of just
10:56trying to get mentioned everywhere.
10:58>> Create video reviews and walkthroughs.
11:00YouTube is one of the most cited domains
11:02by LLMs. So, video has become another
11:05source AI can learn from about your
11:07product. There are a few types of videos
11:08that work well for LLM seeding. Product
11:10walkthroughs that show your tool in
11:12action step-by-step, how-to guides that
11:14solve a specific problem your audience
11:16is searching for, and comparison videos
11:18that put your product side-by-side with
11:20alternatives in your category. You can
11:22also partner with trusted creators in
11:24your space and ask them to produce
11:25independent reviews or demos. That
11:27third-party perspective carries extra
11:29weight because AI tends to treat
11:31independent commentary different from
11:33your own marketing content.
11:34>> Yeah, video content, absolutely agree.
11:36It's something that we recommend also
11:38for all local businesses. Take the most
11:40important service pages or articles like
11:43pricing guides and create useful videos
11:45around that content. Ideally, have the
11:47owner explain the work. If they're
11:48replacing a water heater, show what's
11:50involved, explain what affected the
11:52equipment choice or what made that
11:54installation different. These are things
11:56that they already explain to customers.
11:58We just want to record that and publish
12:00it on YouTube and embed that on the
12:02relevant service page or guide. Someone
12:05reading about the service, not super
12:07common for local businesses, but maybe
12:09you'll have some and they'll see the
12:10actual business doing the work and
12:12explaining it. The quality does not need
12:14to be amazing. We don't need a
12:15production crew, we don't need special
12:17lighting, we don't need scripting. We
12:19just need the person who knows the work
12:21to be able to explain something useful.
12:24As she said, LLMs really enjoy citing
12:27YouTube-based content. So, if you
12:29publish a bunch of content, so if you
12:31publish a pricing guide for water heater
12:33replacement in Gary, Indiana and you go
12:36in depth on that and talk about the
12:38different aspects that affect water
12:39heater pricing. The next time someone in
12:42Gary, Indiana asks, "How much does a
12:44water heater replacement cost?" the
12:46likelihood that your business is going
12:48to be cited because of your video guide
12:50is quite high. Most local businesses
12:54aren't producing video content. This is
12:56a gap that you can exploit by producing
12:58your own.
12:59>> Four, activate social and partner
13:01distribution. Reddit, Quora, LinkedIn,
13:04YouTube, Instagram, and Facebook are
13:05some of the most cited sources by LLM.
13:08They all show up consistently in
13:10AI-generated answers. So, you want your
13:12brand showing up across these platforms.
13:14Have your community managers engage in
13:16relevant Reddit and Quora discussions
13:18where your audience is asking questions.
13:20Work with your social media team to
13:22share product insights on LinkedIn.
13:24Partner with Instagram influencers in
13:26your space who can talk about your
13:27product to their own audiences. Your
13:29employees can also play a big role here.
13:32When your team members share their
13:33expertise and mention your product
13:35naturally on their own profiles, that
13:37adds another layer of real authentic
13:39presence that AI can pick up.
13:41>> Okay, so now we're talking about Reddit,
13:43Quora, LinkedIn, YouTube, Instagram, and
13:45Facebook. This is a lot of work for a
13:47small plumber in Gary, Indiana. There
13:49are markets where a bigger distribution
13:51push makes sense. If you are in Houston,
13:54for example, Houston is very competitive
13:56and a business targeting that market
13:58probably has the budget to support some
14:00of this work. But most small businesses
14:02need to prioritize. I talked about
14:04YouTube. Beyond that, I generally will
14:07focus on useful content on their website
14:10before committing to all of these
14:11different platforms. Again, water heater
14:13replacement pricing. The owner can
14:15explain what they charge, what's
14:16included, what changes the price. A
14:19straightforward replacement is one job.
14:22An installation with difficult access or
14:24additional work is another. Explain
14:27those differences using jobs that this
14:30business actually handles. And of
14:32course, the local requirements that
14:34actually apply and adjust so that
14:36someone researching the cost has useful
14:38information from a business that can do
14:40the work in their area. And AI systems
14:43have specific information that they're
14:44going to be able to reference for a
14:46relevant local question about this. This
14:49is what we call trust content and I want
14:51to show you that we mean this. So if I
14:53come into the Court 38 Agent, this is
14:55the the software that we use that
14:57researches and writes local content
14:59designed to build trust and relevance
15:01for Google Business Profiles and for AI
15:04systems. One of the types of content
15:06that we have here is trust. So we can
15:08select which services we want to write
15:10trust content for. We can check to make
15:12sure the writing agent has the necessary
15:15information. We can check to make sure
15:17it has the necessary pricing information
15:19and then we can simply have it create
15:22the pricing guide for that relevant
15:24service. This is very valuable. So few
15:26local businesses are transparent with
15:29their pricing. This means that the LLMs,
15:32when they're asked, "How much does this
15:34service cost?" Uh they want local
15:37results. They want local information.
15:39Google does, too. If none of your
15:41competitors have published pricing
15:43guides, and you go and price it, very
15:46high likelihood that you're going to get
15:47cited. So, for these smaller local
15:49businesses, this type of content is so
15:52much more important than like a month of
15:55Instagram posts. Uh get the important
15:58customer questions answered properly
16:01first.
16:02>> Encourage detailed customer reviews.
16:04Real user feedback is one of the
16:06strongest signals for AI because it
16:07comes from people describing actual
16:09experiences with your product. So,
16:11encourage your users to leave detailed
16:13reviews on platforms like G2 and
16:15Capterra. But, don't just ask for a star
16:17rating. Prompt them with very specific
16:19questions. Something like, "What
16:20specific features helped you solve a key
16:22problem?" Or, "How does this product
16:24compare to others you've used?" That
16:26level of detail gives AI models much
16:28clearer context about what your product
16:30actually does and who it works best for.
16:32A review that says, "Great tool, five
16:34stars." doesn't help AI understand
16:36anything. A review that says, "This
16:38saved our three-person team 10 hours a
16:40week on invoicing." gives AI something
16:42concrete to reference.
16:43>> Yeah, completely agree with this. And in
16:45fact, the way that we coach our clients
16:47to request reviews has changed because
16:50we know that the AI systems are reading
16:52all of the reviews, right? Google
16:54actually released an example saying if
16:56somebody goes into Google Maps and says,
16:59"Find me a coffee shop with fast Wi-Fi."
17:01it's going to read all of the reviews of
17:03coffee shops around you, and if one of
17:05them mentions fast Wi-Fi in a review,
17:07that one is very likely to get cited, to
17:10get recommended. Now, we used to coach
17:13our clients to just text their prior
17:15customers and say, "Hey, would you mind
17:17leaving us a review?" and then dropping
17:19the Google review link. That's not what
17:21we do anymore. Now, we coach our clients
17:23to text their customers and say
17:25something more detailed. Could you
17:26please share what we helped you with and
17:28what the outcome was? This follows all
17:30of Google's current rules around review
17:33solicitation, but is much more likely to
17:35get the user to talk about what
17:38happened, to talk about their
17:39experience. And that's where we get
17:41attributes mentioned in these reviews,
17:43right? Maybe the customer can mention
17:45clear pricing. Maybe they value how the
17:47technician explained options. These are
17:50details that the AI systems can use to
17:52understand your business and describe it
17:54effectively to people who are asking
17:56them for a recommendation. We also want
17:58to use information in these user reviews
18:01to actually write the content on your
18:03website. We do this in a couple of
18:04different ways. Every time the Court 30
18:07Agent writes an article, it's going to
18:09read every review about your business
18:11and every review about your competitor's
18:13business. The information it gets about
18:16your business reviews is going to
18:17incorporate into the content that it
18:19writes about your business. The
18:20information on competitor reviews it's
18:23going to incorporate into various pain
18:24points or issues that customers in your
18:27local area have had with that exact
18:30service. And to get even more
18:32information out of the owners, we
18:34actually built an audio interview
18:35process into the Court 30 Agent itself.
18:38It happens when you're creating a new
18:40business in the agent itself, and it
18:42goes through, makes sure that your
18:45settings are correct, and then it starts
18:47actually asking you questions over your
18:49speakers and allows you to answer
18:51through your microphone. We found that
18:54if we send a questionnaire to our
18:55clients, we typically get very short,
18:57uninteresting responses. But if we get
18:59them talking about their business, this
19:01is where we can get real information to
19:04write unique content that's only
19:06applicable to their business. We want
19:08our content to feel like only this
19:11business, only in this location could
19:14that content be applied to them. That's
19:15how we make sure that we're writing
19:17content that meets Google's
19:19informational gain guidelines.
19:21>> Those are the five steps to put LLM
19:22seeding into action. Now, if you already
19:25have an outreach strategy for backlinks
19:26and PR, you're not starting from scratch
19:28here. LLM seeding adds a new layer to
19:30that same work. The difference is that
19:32now you're also and specifically
19:34thinking about how AI discovers,
19:36understands, and references your
19:37product. That takes intentionality. You
19:39have to think about what to publish,
19:41where you place it, and how consistently
19:43your product is described across every
19:45source. And remember, this is not a
19:46one-done effort. The third part of the
19:48framework is reinforce. That means you
19:50keep showing up consistently across all
19:52your channels so AI continues to
19:54encounter your brand over time. So, keep
19:56updating your landing page as your
19:58product evolves. Keep publishing fresh
20:00content that reflects your latest
20:01features and use cases. Revisit your
20:04third-party listings to ensure the
20:05information is still accurate. Ask for
20:07new customer reviews as you release
20:09updates or serve new audiences.
20:11>> When we measure this at my agency, we
20:13generally focus on tracking the Google
20:16Maps rankings across a rank map. That
20:18tells you where you're ranking if
20:20someone searches for that specific
20:22service entity in a very specific
20:24location. So, a green ranking map tells
20:27you the business ranks in maps. Now,
20:29unfortunately, that doesn't tell us
20:30whether Google actually recommends it in
20:32the AI overview. Tracking
20:33recommendations in AI is incredibly
20:36difficult right now. I know there are
20:38dozens and dozens of tools that claim to
20:40do this. I have yet to find one that can
20:42do it accurately. There are a few
20:44reasons for this. Number one, building a
20:46tool capable of querying uh an API with
20:50a specific prompt and then seeing what
20:52comes out of it is something that can be
20:53done in roughly 45 minutes with Claude
20:56code, meaning that these AI tracking
20:58tools can be built very, very quickly
21:00and easily. The problem is they just
21:03don't work very well. There's a lot of
21:05evidence that recommendations that
21:07results coming out of an API query are
21:10very different than recommendations that
21:13come out of somebody who's using the
21:14desktop app. This means that unless you
21:16think your local customers who need a
21:18plumber are firing up Claude code to do
21:21an API inquiry, the API results are not
21:24very applicable to whether you're being
21:27recommended or not. Now, there are some
21:29tools that can effectively track using
21:31the desktop app. Those are generally
21:32harder to do. They're more expensive,
21:34but even those have issues. The problem
21:37comes with how hyper specific, how much
21:40each agent, each LLM learns about the
21:43individual user. So, if I ask the LLM
21:46something and then you take that exact
21:47same prompt and put it in, you're going
21:50to get different recommendations. On top
21:52of that, if you take the same prompt and
21:54put it into chat GPT three or four
21:56different times, you also will very
21:58often get different recommendations. So,
22:01what we generally do is track AI
22:04referral traffic in Google Analytics. We
22:07just look for the trend. Is it going up
22:09or is it going down? The reason we're
22:11just looking for the trend is because
22:12the number is a vast undercount. Most of
22:15the time when an LLM recommends a
22:17specific business, the user isn't
22:19clicking through to land on that
22:21business's website. The user is then
22:24searching on Google for that brand name
22:26to find the business. Okay? So, Google
22:29Analytics gives you an idea if it's
22:31getting better or worse, but
22:33individually tracking services or
22:36service entities very, very difficult to
22:39do accurately, so we usually don't. At
22:41the end of the day, I'd rather not have
22:43data than have bad data that leads me to
22:46an inaccurate conclusion. For the
22:48website structure, if you want to dive
22:50into that and know exactly how we
22:51structure a website for a local business
22:54to make sure that LLMs can read and
22:56understand your business, uh check out
22:59this video next.

About this transcript

This page contains the full transcript of Save Your Business from AI (Do This!) by Caleb Ulku, generated from the public captions YouTube serves with the video. The transcript has 4,219 words across 660 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.