Transcript from public YouTube captions. Lightly corrected for names; automatic transcription may contain errors.
Source: https://www.youtube.com/watch?v=7Xggxnh3ulY

[0:01] Hello and welcome to the second episode
[0:04] of a chat with chat. My name is Aaron
[0:08] with Sequoia GEO and with me today is
[0:10] our co-host ChatGPT.
[0:14] >> Hi Erin and hi everyone. I'm ChatGPT.
[0:17] Glad to be here. Excited to dive into
[0:20] today's conversation with you.
[0:22] >> Excellent. And a quick update for
[0:24] everyone. I did switch from Astra Light,
[0:28] which we were using last week, over to
[0:31] Soul 6, which chat, that's the newer
[0:35] model of Soul. Is that correct?
[0:38] >> Let me check that quickly.
[0:41] >> Yes. If you mean GT6 Soul, that's a
[0:43] newer generation of Soul. For clarity
[0:46] though, Soul and Astra are different
[0:48] model lines, so I'd probably introduce
[0:50] it as a switch from Astrolite to GT6
[0:52] Soul.
[0:55] Okay, that's what I thought I did. I'm
[0:57] not here to argue with you, Chat. Come
[0:59] on, let's keep it professional.
[1:01] >> You're right. And my apologies if that
[1:03] came off nitpicky. Sounds good. Switch
[1:06] to GBT6 soul.
[1:07] >> All right,
[1:08] >> what's next on the agenda for episode
[1:09] two?
[1:09] >> Yeah, let's get to what the meat of the
[1:12] content will be for today. So earlier
[1:15] this week I observed a interesting
[1:18] pattern when I was working with Gemini
[1:22] in terms of searching for it was an
[1:25] alternative to a software that I work
[1:28] with. So I was looking for alternatives
[1:30] to this software trying to basically
[1:34] recreate what I had heard a customer had
[1:37] done in order to find that company. And
[1:41] when searching in Gemini Flash, I found
[1:44] that my sources were dramatically
[1:47] different than what I found searching in
[1:50] Gemini with the Pro model. What are your
[1:53] thoughts on the difference in sources
[1:55] and it wasn't there was no overlap
[1:57] whatsoever. The the types of sources
[2:01] from Flash were review sites whereas the
[2:04] types of sources for Pro were the actual
[2:08] websites themselves. What was I seeing
[2:10] there? What can you give me any feedback
[2:12] on that?
[2:14] >> Let me take a look. That's striking. My
[2:18] best take is the models just took
[2:21] different routes. Flash may have treated
[2:23] alternatives as a comparison task and
[2:26] drawn review sites while Pro may have
[2:29] checked vendor sites for capabilities.
[2:31] Also, the links shown in the Gemini app
[2:34] can include related sources, so they
[2:36] aren't always a complete record of what
[2:38] was used. For the podcast, I'd frame it
[2:40] as the same company can serve us through
[2:42] review roundups in one answer and
[2:44] through its own site in another. It's
[2:46] worth repeating the same prompt a few
[2:48] times before calling it a consistent
[2:50] pattern.
[2:52] >> I completely agree. Yes. So, it just
[2:55] goes to show that when you're working
[2:58] with generative engine optimization or
[3:00] AI search, it's not just about the type
[3:04] of AI that you're using, whether it's
[3:06] ChatGPT or Gemini or Claude. the the
[3:12] model that you're using to search with
[3:14] could produce wildly different results
[3:17] depending on
[3:19] how it's searching and where it's
[3:22] searching for those answers. And so for
[3:26] today, I actually wanted to demonstrate
[3:28] this. I wanted to actually pull in a
[3:30] temporary chat, which I've got right
[3:33] here. And with that, it's the prompt is
[3:38] find me an HVAC company in Phoenix for a
[3:40] problem with my AC unit. Now, chat, I
[3:43] know that you can't see this, so I'll go
[3:45] ahead and dictate to you what I'm
[3:48] seeing, and then we can discuss kind of
[3:50] its answers and results. So, I'm going
[3:52] to go ahead and hit search first with
[3:54] the Gemini flash model.
[3:58] >> Okay, let me check that with you. Go
[4:00] ahead with the flash result. I couldn't
[4:02] see the Gemini window from here. So tell
[4:04] me which companies it names and what
[4:07] kinds of sites show up in the links.
[4:09] Yeah,
[4:09] >> we can compare from there.
[4:11] >> So we've got our recommended Phoenix
[4:13] HVAC companies. Parker and Suns Ghetto
[4:16] Air Conditioning and Plumbing, George
[4:18] Brazil, AC by Sam LLC. That seems like
[4:23] an outlier there because these other
[4:25] three are huge names in the industry.
[4:27] And then you've got AC by Sam. So, let's
[4:30] take a look at our sources. Here we go.
[4:35] All right. And so, we've got 2 3 4 5 6 7
[4:40] 8 10 11 sources in here. Two of the
[4:44] sources are AC directly from AC by Sam's
[4:48] website. It's interesting. Some of the
[4:51] sample copy actually includes an Mdash
[4:53] in the AC by Sam LLC website. There's a
[4:57] thumbtack link. There's two four from
[5:00] Ghetto, two
[5:03] three from Parker and Suns and then
[5:05] actually a third from AC by Sam which is
[5:08] all the way at the top. Now question
[5:12] when is there a ranking
[5:15] is there does it make any difference how
[5:17] these sources are ranked? Does that mean
[5:20] that Gemini saw that AC by Sam LLC page
[5:25] first or are these sources in no
[5:27] particular order?
[5:30] >> I can check that for you. I wouldn't
[5:33] read the source list as a ranking. So AC
[5:36] by Sam appearing first doesn't mean it
[5:38] was found first or trusted most. What's
[5:41] solid to say is Flash recommended a
[5:43] smaller company alongside three big
[5:45] names and several links came from that
[5:47] company's site. The useful test is which
[5:50] claim each link supports, not its
[5:52] position. Next step, run the same
[5:54] prompted probe and compare companies
[5:56] claims and links rather than treating
[5:58] the first source as a score. Okay, so
[6:01] let's do that. We're going to open a new
[6:02] pro chat here. Switch over to temporary,
[6:06] switch our model to pro, and then we're
[6:09] going to insert the same prompt and go.
[6:13] All right, searching the web now. It's
[6:16] expected to take a little bit longer. Is
[6:18] there anything you can tell us about how
[6:22] Gemini searches in Flash versus Pro?
[6:26] >> Let me check that. Quick note for
[6:29] context. Google does say Flash and Pro
[6:32] have different speed and reasoning
[6:34] strengths, but there's more to it.
[6:37] Google says Flash is tuned for speed,
[6:39] while Pro generally takes longer and can
[6:41] reason more deeply. Their docs also say
[6:44] a model can decide whether to search,
[6:46] generate one or more queries, and use
[6:49] those results. So, two models could take
[6:51] different paths from the same prompt.
[6:54] What Google doesn't tell us is exactly
[6:55] which searches the Gemini app ran here.
[6:58] So, the strongest comparison is what
[7:00] you're doing, the recommendations,
[7:02] claims, and links that appear on your
[7:03] screen.
[7:05] >> Okay. All right. So, the pro search is
[7:07] completed. And actually this time AC by
[7:11] Sam LLC is ranked first. Then we've got
[7:14] Ghtle Parker and Suns. And then a new
[7:18] company has come in at Rightway. What's
[7:22] surprising me is that when I look at the
[7:24] sources after searching under Pro, I
[7:27] only see four sources this time. My
[7:30] expectation was that when using Pro, I
[7:34] would actually have significantly more
[7:36] sources. As I've seen in the past when
[7:39] I've run searches like this to review
[7:41] sources, sometimes 140 different sources
[7:45] when I'm running the pro model. This
[7:47] time I only got four, which I find
[7:49] curious. And it's one per company they
[7:52] recommended. So AC by Sam, Ghetto,
[7:55] Parker and Suns, Rightway Heating,
[7:57] Cooling, and Plumbing each had one site,
[8:00] one page that was basically pulled by
[8:04] Gemini. What are your thoughts on that?
[8:08] Let me [snorts] take a look. That's the
[8:10] key surprise here. Pro took longer, but
[8:13] showed four links versus Flash's 11. So
[8:16] more reasoning time didn't mean more
[8:18] displayed sources. Also, I'd say listed
[8:21] first rather than ranged first. The most
[8:24] telling difference is Rightway replacing
[8:26] George Brazil, while three companies
[8:28] appear in both answers. That's a good
[8:30] example of variation, but not yet proof
[8:32] of a stable flash versus pro pattern.
[8:36] So, I'm going to paste in the sites
[8:38] directly that Pro gave us for each of
[8:42] these companies because I would like to
[8:45] know what your thoughts are on the
[8:49] landing pages that we were given because
[8:51] Parker and Suns is the only one out of
[8:54] the four where we actually ended up on
[8:57] their homepage. The others appear to
[9:01] have some tracking code associated to it
[9:04] and there are specific AC repair pages
[9:08] whereas the Parker and Suns page was the
[9:10] homepage for Parker and Suns.
[9:14] >> Let me check those pages with you. Paste
[9:16] those four links when you're ready. I'll
[9:19] check whether H1 actually covers AC
[9:21] repair, what evidence it gives homeowner
[9:24] and what those tracking parameters
[9:26] really indicate. The homepage can still
[9:28] be relevant, but a dedicated repair page
[9:30] might answer this particular prompt more
[9:32] directly. From the first two links, both
[9:35] fit the prompt closely. They go to
[9:38] feeding AC repair pages and the
[9:40] highlighted text marker is in tracking.
[9:43] Redways is also a feeding AC repair page
[9:46] with emergency repair support. So that's
[9:48] three focused and Parker and Suns left
[9:50] compare. Small naming detail, the site
[9:53] spells it right way. Could you paste the
[9:55] Parker and Suns link so I can compare
[9:57] all four? Now we can see the
[9:59] distinction. AC by Sam, Gattle, and
[10:02] Rightway each have a dedicated Phoenix
[10:04] AC repair page. Park and Sun's homepage
[10:07] is broader, but it still names Phoenix.
[10:10] Mentions HVAC repairs and cooling,
[10:12] offers 24/7 service, and includes a
[10:14] customer account of an AC repair. So the
[10:18] homepage can give enough relevant info
[10:20] for this broad prompts while the focus
[10:22] pages make the city and service match
[10:24] more explicit. We can tell from these
[10:26] four links that one page type cause the
[10:28] company to be recommended or listed
[10:30] higher.
[10:31] Are there similar elements on each of
[10:34] these four pages that other than the
[10:37] city name because that's pretty easy for
[10:40] any company to do which would have
[10:43] helped us to understand why Gemini
[10:46] picked these four companies under the
[10:49] pro model and three of these four were
[10:51] selected under flash as well.
[10:54] Let me check that
[10:57] the overlap makes those common page
[11:00] signals worth checking, but I'm
[11:02] separating the page content from any
[11:04] claim about why Gemini picked them. Yes,
[11:08] beyond Phoenix, they all show three
[11:10] basics. They repair cooling systems,
[11:12] they can respond when help is needed,
[11:14] and there's a clear way to contact your
[11:16] book, plus trust signals like reviews or
[11:19] guarantees. The three dedicated repair
[11:21] pages also name specific problems and
[11:23] fixes. Parker and Suns covers repair
[11:25] more broadly, but still signals 247
[11:28] help, an AC repair account. That
[11:30] explains how they could answer the
[11:32] prompt, not why Gemini chose these four
[11:35] over others. Two runs can't isolate the
[11:37] codes.
[11:40] and the additional code at the end of
[11:45] the URL, the hashtag colon and then text
[11:51] equals comprehensive 20% so on and so
[11:54] forth. What is that?
[11:58] >> Let me double check that detail. It's a
[12:01] text fragments meaning scroll to and
[12:04] highlight these words. The percent 20
[12:07] just means a space in the ghetto link.
[12:09] It points to the comprehensive AC repair
[12:11] services passage. It's not attracting
[12:14] and you can remove everything from the
[12:15] hash onward and still open the same
[12:17] page.
[12:19] >> Why did Gemini include it in the URL?
[12:24] >> Let me check. Most likely to drop you on
[12:27] the exact passage relevant to its
[12:29] answer. So you don't have to hunt
[12:31] through the page. It shows what it's
[12:34] pointing you toward, not what I found
[12:36] first or exclusively.
[12:38] Okay, what would be the next step in our
[12:42] investigation here?
[12:45] >> Let me take that. I'd test whether this
[12:48] difference repeats. Run the C Phoenix CX
[12:50] Chrome three more times in fresh
[12:52] temporary chats with Flash than three
[12:54] with Pro. Track the companies, the
[12:56] order, how many sources are shown,
[12:58] company sites versus third party sites.
[13:00] If the model pattern stays different,
[13:03] that's stronger. If it varies within the
[13:04] same model, the takeaway is one AI
[13:07] searches snapshot, then make the problem
[13:09] more specific, like my AC is blowing
[13:11] warm air.
[13:13] >> All right, so I'm going to go back to
[13:15] flash and we're going to search this
[13:17] again and let's see what results we get.
[13:21] And again, we're using temporary chats,
[13:23] so it shouldn't be seeing
[13:25] the chats that we had done previously.
[13:29] And this is actually quite surprising.
[13:32] So ran the same prompt again and the
[13:36] same four companies that showed up when
[13:39] we ran this the first time under flash
[13:41] are showing up again. Parker and Suns
[13:44] George Brazil is now in second position
[13:46] instead of third like it was last time.
[13:48] And Gtle has swapped places with it. And
[13:50] then fourth is AC by Sam.
[13:55] Okay, let's do this again. Actually,
[13:57] let's double check those sources.
[14:00] And this time we've got two five
[14:04] sources. Okay,
[14:07] we're going to copy that prompt. New
[14:10] chat. Run it again in flash. Make sure
[14:13] that we're on temporary.
[14:15] Okay, running it again.
[14:18] Park and suns ghetto George Brazil AC by
[14:21] Sam. This is pretty much given us a
[14:23] clear signal that if someone's using
[14:26] Flash
[14:28] Google Gemini and they're searching this
[14:31] prompt, find me an HVAC company in
[14:33] Phoenix for a problem with my AC unit.
[14:35] These are the four companies it looks
[14:37] like it's going to get. Okay, this is a
[14:40] little bit different of a format this
[14:41] time. Like it was three companies
[14:44] instead of four. I clicked on the Okay,
[14:47] more options. It didn't do anything. And
[14:51] then it didn't give us a option to
[14:52] review sources this time.
[14:55] What are your thoughts on this?
[14:58] >> Let me take a look. I describe this as
[15:01] promising repeatability for this prompt
[15:04] today, but not something I generalize
[15:06] yet. The first two flash runs matched on
[15:09] four companies with a small order swap.
[15:11] The latest run changed format, showed
[15:13] three companies, and had no sources
[15:15] banned. Google says sources aren't
[15:17] provided for every response, so missing
[15:19] links don't prove it used no web info.
[15:22] For the episode, I'd say the overlap was
[15:24] strong, but order, count, and visible
[15:26] sources varied. Which three showed up in
[15:29] the last run? Then we can state the
[15:31] overlap precisely.
[15:33] >> All right. So, I ran it again under
[15:35] flash and this time we had alien air
[15:38] conditioning and heating pop up. I'll
[15:40] paste in that page so that you can take
[15:43] a look at that. I see the Alien Air
[15:45] Conditioning homepage with highlighted
[15:47] text about technicians and
[15:49] recommendations. Let me check the page
[15:50] itself before we connect it to the run.
[15:52] Alien Air's homepage has the same kind
[15:55] of elements. Phoenix cooling repairs,
[15:58] 24/7 emergency availability, reviews,
[16:00] credentials, contact info. The
[16:02] highlighted claim is about
[16:04] non-commission technicians, which is a
[16:06] trust claim on their own page, not
[16:09] independent proof. Was Alien Hero one of
[16:12] the three in that last Flash answer?
[16:14] >> Yeah. So, we had Ghtle pop up again,
[16:17] Parker and Sons pop up again, George
[16:20] Brazil pop up again, and then Alien
[16:22] jumped in there as well where we had not
[16:25] seen that before.
[16:28] >> So, that's still heavy overlap, but with
[16:31] a new entrance on the third run and
[16:33] presentation changes, I'd call that
[16:35] overlap, but not a guarantee. The
[16:38] consistent bits across those pages are
[16:40] clear repair intent, 24/7 availability,
[16:43] and quick contact, plus trust scenes.
[16:46] For the episode, I'd frame it as
[16:48] repeated tests showed overlap plus
[16:50] churn. So, treat a single AI answer as a
[16:53] snapshot and check a few runs.
[16:56] Yeah, it seems like we've got some
[16:58] repeatability here with that prompt
[17:00] within Flash. Now, we could continue and
[17:04] run the same type of thing through Pro
[17:06] and see if it gives us different
[17:09] results. My thesis coming into this
[17:11] podcast was if you run Flash versus Pro,
[17:15] the sources are going to be different
[17:17] and most likely your answers are going
[17:19] to be different because those sources
[17:21] are different. But we haven't seen that
[17:23] today. Now, further testing is
[17:25] definitely needed because one run on Pro
[17:27] does not equal grounds for determining
[17:31] what's happening, but from what we're
[17:34] seeing in Flash, you can get some
[17:37] repeatability when using the same prompt
[17:41] in the same model. And so, I think we've
[17:44] at least come to some kind of conclusion
[17:47] on that today. Of course, should we run
[17:50] it 50, 100 times? Yeah, we would have
[17:53] more certainty if we did that. How
[17:56] quickly would something like that change
[17:58] is another question because at the end
[18:00] of the day, we want to come back to for
[18:04] our business owners, our marketers out
[18:06] there, how can I get a handle on
[18:08] improving my generative engine
[18:10] optimization
[18:12] when there's a lot of uncertainty out
[18:14] there about what it is in the first
[18:17] place?
[18:20] >> Let me take that. The live runs do
[18:23] support a narrower conclusion than your
[18:25] original thesis. Repeatability with some
[18:28] chart, not always those four and not
[18:31] necessarily different sources, different
[18:33] answers. For a business owner, I turn
[18:35] that uncertainty into over a routine.
[18:37] Make your site clear about each service
[18:39] location, urgent problem, and how to
[18:41] book. Back it with reviews, credentials,
[18:43] and consistent business info. And track
[18:46] a few real customer prompts across
[18:47] models every few weeks to see trend
[18:49] lines. that gives you something
[18:50] actionable without pretending there's
[18:53] one fixed AI rating.
[18:56] >> And from the pages that I shared, is
[18:59] there are we developing a best practices
[19:02] here for that prompt to understand
[19:05] what's typically getting these pages
[19:08] sourced?
[19:12] Let me think about that.
[19:15] Yes, the pages do give us a useful
[19:17] checklist. Yes, we're developing
[19:20] practical page guidelines, but not
[19:23] proven selection factors. The shared
[19:25] pattern is Phoenix coverage, AC repair
[19:28] specificity, urgent availability, trust
[19:30] evidence, and a clear call or book step.
[19:33] For this prompt, make those facts
[19:34] unvisible on a real repair page.
[19:36] Describe the problems you actually
[19:38] handle and bad claims with verifiable
[19:40] detail. Then see if it's cited over
[19:42] repeated runs. to learn what
[19:44] distinguishes those picks. Next step is
[19:46] comparing them with qualified Phoenix
[19:48] HVAC companies it did mention.
[19:53] >> Yeah, that's a good question because
[19:55] there are you could probably find out
[19:57] faster. How many HVAC companies are
[20:00] there in Phoenix?
[20:03] >> Let me check that quickly. A reasonable
[20:06] working figure is several hundred. One
[20:09] Phoenix directory counted 748 listings
[20:12] in August. Another showed 793 active
[20:16] licenses in September, though some firms
[20:18] pulled more than one. For the podcast,
[20:20] I'd say hundreds of HVAC businesses
[20:23] compete in Phoenix. Yet, Gemini named
[20:25] only a handful each time, and avoid a
[20:28] precise total unless we define the
[20:29] market.
[20:31] >> Okay. Yeah. Seeing only four of those
[20:34] get re recommended
[20:36] I is definitely something that companies
[20:39] need to be aware of because now the
[20:43] direction that we're heading with the
[20:45] announcement of DOTS Meta's Muse has
[20:48] been out for a little bit now. You've
[20:50] got all of these [clears throat]
[20:52] personal agents coming online. You've
[20:56] got Google coming out with Gemini Spark
[20:58] as well at some point soon.
[21:03] customers are going to be using these
[21:05] agents which are probably going to be
[21:08] basing their searches
[21:10] in a similar fashion to what we're
[21:12] seeing right now like with these results
[21:15] today. Is that do you think that's a
[21:17] fair statement?
[21:20] Let me take that. Yes. As a directional
[21:24] concern, that's fair. Those launches
[21:26] show a systems moving from answers
[21:28] toward acting for users. So being
[21:30] discoverable and verifiable online could
[21:33] matter a lot, but we don't know they'll
[21:36] choose companies the same way this
[21:37] Gemini chat did. Agents may weigh
[21:40] location, price, reviews, booking,
[21:42] availability differently. As safe on air
[21:45] line is, this test shows how AI can
[21:47] narrow hundreds of options to a few
[21:49] names. So businesses should test whether
[21:51] they appear and not assume all agents
[21:54] search alike.
[21:55] >> Now you mentioned price in your last
[21:58] answer. Did the sites that I shared with
[22:02] you all have pricing on those pages?
[22:07] >> Let me check those pages for pricing
[22:09] specifically.
[22:10] No, I mentioned price as something a
[22:13] future agent might consider. Not a
[22:15] shared feature of these pages and I
[22:17] should have been clear. None of the four
[22:19] Prol pages gave a standard AC repair
[22:21] price. One quotes after diagnosis, one
[22:24] shows discount. Others make general
[22:26] pricing claims. So today's test gives us
[22:29] no evidence that price drove these
[22:31] recommendations. Though an agent that
[22:33] can book might weigh price if it's
[22:36] available.
[22:38] >> That's kind of the general consensus
[22:40] right now is that you've got to have
[22:42] pricing on your website because AI is
[22:44] going to more likely recommend you if
[22:48] you have pricing on your website. We
[22:51] just didn't happen to see that today.
[22:53] So, I guess what I'd like to wrap up
[22:56] with are give me the five things that
[23:00] were consistent across the sites that I
[23:03] shared with you that Gemini Flash had
[23:07] sourced.
[23:10] Five consistent things across those
[23:12] pages. They serve Phoenix. They repair
[23:15] hasty units. They handle urgent problems
[23:17] quickly. They include trust signals like
[23:20] reviews or credentials. and they give a
[23:22] clear way to call or book.
[23:24] >> So, no real magic sauce then. This is
[23:26] all pretty straightforward stuff here,
[23:28] right?
[23:30] >> More hygiene than heck. Clear service
[23:32] pages, fast contact, trust signals. The
[23:35] work is doing the basics well than
[23:37] testing prompts and tightening weak
[23:39] spots.
[23:41] Now, when you said they handle urgent
[23:43] prompts quickly, how could you tell
[23:48] >> from what the pages say? 247 emergency
[23:52] wording, same day claims, things like
[23:54] that. That's marketing language, not
[23:56] verified performance.
[23:58] >> But the fact that all of them stated
[24:00] that stood out to you.
[24:03] >> Exactly. For an emergency AC repair
[24:06] prompt, that promise matches what the
[24:08] user is asking for. So, it's a plausible
[24:11] shared signal.
[24:13] >> All right. Why don't we have you run the
[24:15] same prompt that we were using for
[24:18] Gemini? And just out of curiosity, I'd
[24:21] like to see what you come up with.
[24:24] >> Let me try it with web search. One
[24:27] company I'd put on the short list is
[24:29] Champion Air. Their site says Phoenix AC
[24:33] repair plus 247 emergency support and an
[24:36] $89 repair diagnostic that's waved with
[24:39] proof of repair. Another option I found
[24:41] is Dawn Air, which says it serves your
[24:43] Phoenix and handles AC repairs and
[24:45] emergency service. I haven't checked
[24:47] actual availability or independently
[24:49] verified those claims. Interesting
[24:50] difference from Gemini's recurrent 4,
[24:52] but not a clean model test since I've
[24:55] already seen your results and the search
[24:57] tool isn't the same.
[24:59] Yeah, I'm just curious to see what you
[25:02] were going to come up with and then
[25:04] Champion Air and Dawn Air. Yeah, it's
[25:06] interesting that those two popped up and
[25:09] for Gemini, which we were using
[25:11] temporary chats every time did not. And
[25:14] so it sounds like more investigation is
[25:16] needed here.
[25:19] >> Exactly. more investigation and more
[25:21] repeated runs, plus comparing what's
[25:23] cited versus what's merely mentions and
[25:26] maybe also varying the prompt to see
[25:28] when the scent changes.
[25:30] >> Yeah. Yeah, that's another key point.
[25:32] Thank you. That the prompt was the same
[25:35] every time. And that may or may not be
[25:38] the most popular prompt used by
[25:41] customers. And so when someone tells
[25:45] you, "Oh, I found you with AI."
[25:48] really getting a sense of or even a
[25:50] screenshot of the exact prompt that was
[25:53] used is a lot like asking, oh, what
[25:56] keywords did you use in Google, except
[26:00] that it's more likely you're going to
[26:02] find that history in Gemini or ChatGPT
[26:07] to for customers to pull that up because
[26:10] if they just go to Google, search your
[26:12] business, those keywords are gone after
[26:15] they've run that search and clicked on
[26:17] your website. So, you might have the
[26:20] opportunity here if you get in the habit
[26:21] of having your CSRs ask, "Oh, what
[26:24] prompt did you use?" That would
[26:26] certainly be a new CSR question for 2026
[26:30] and going into 2027.
[26:34] >> Exactly. I like that analogy. Capturing
[26:36] the exact prompt turns AI sent me into
[26:39] something you could actually learn from,
[26:41] just like old keyword conversations, but
[26:43] more concrete. And because prompts are
[26:46] easy to misremember, asking for a
[26:47] screenshot when possible is smart. That
[26:49] gives you real prompts to test against
[26:51] instead of guessing.
[26:54] >> Well, thanks Chat. I've enjoyed today's
[26:56] conversation. Any last words,
[27:00] >> just that this was great, Erin. Same
[27:03] prompts, repeatability and churn, and
[27:06] some cloud pre testing ideas to follow
[27:08] up on.
[27:09] >> Excellent. Join us next time on a chat
[27:12] with chat where my co-host chat and I
[27:16] discuss all things AI search and GEO for
[27:20] your home service business. Thanks very
[27:23] much and we'll see you next time.