Randall Kanna on AI Content Optimization

00:00 Hey, Tim Berglund here. I recently had

00:01 the pleasure of talking

00:02 to Randall Kanna. Randall

00:04 is an author, engineer, and teacher who's

00:06 working on a really

00:07 cool startup idea that

00:08 got my interest. The way I see it, nobody

00:11 really knows right

00:12 now the best way to get

00:13 our content noticed by AI models. It's

00:16 kind of like the dark

00:17 ages of SEO all over again.

00:20 This matters for anybody writing or

00:22 making a video or posting

00:23 anything online, but it's

00:25 of keen interest to people working in

00:27 DevRel. So naturally, I

00:28 wanted to learn what I could

00:29 from Randall, and it turns out that was a

00:32 lot. So let's listen in.

00:35 Randall, welcome to the show.

00:37 Thanks for having me.

00:38 You bet, thanks for being here.

00:40 tell us a little bit about yourself.

00:41 What's your professional journey?

00:43 How did you get to this moment in time?

00:46 I know there are a lot

00:47 of folks who know you,

00:48 but for some folks who

00:49 don't, tell us about you.

00:50 I'm sure there's a lot of

00:51 folks who don't know me,

00:51 so say most of them don't know me.

00:54 But I appreciate the intro and again,

00:56 thanks for having me.

00:57 It's actually my first

00:58 podcast since my second son

01:00 was born, so it's been a while for me.

01:02 I've been an engineer

01:03 for over 10 years now.

01:05 I moved to San Francisco in my early 20s,

01:08 went to a coding bootcamp.

01:10 I started the coding bootcamp already

01:12 knowing how to code,

01:13 which was probably in retrospect,

01:14 I probably could have saved some money

01:16 and just tried to apply to jobs.

01:20 But I was so nervous to attend.

01:21 I was so nervous to

01:23 attend such imposter syndrome

01:26 that I taught myself

01:27 and deployed several apps

01:28 before I got there.

01:29 And went there, got a job 10 days after,

01:33 got I think two offers

01:35 within the first two weeks,

01:36 signed an offer very quickly.

01:39 I had just moved back to my

01:40 parents' house in Sacramento

01:42 and I was staying with them

01:44 rent free for about 10 days.

01:47 Then I immediately

01:47 moved back, told the job,

01:48 I already lived there.

01:50 And signed a lease, signed the offer.

01:52 We did not too long of a drive though.

01:53 Yeah, it wasn't too bad.

01:55 But I had an apartment immediately

01:57 and then I've worked since then,

01:58 I've worked at large

01:59 companies such as Eventbrite,

02:02 Ticketfly, Pandora,

02:03 I'm an O'Reilly author.

02:05 I've self published two of my own books,

02:07 launched an engineering course

02:08 and sold a very small app, Code Tutor.

02:12 And now I am launching Kelsey.

02:14 Tell us about Kelsey.

02:16 And this is the thing, I saw you talk

02:18 about it on Twitter.

02:20 I had recently written about the topic

02:24 and by written about the topic,

02:25 I mean there was like a paragraph

02:26 or maybe a sentence that said,

02:28 "Hey, this matters."

02:30 So super well informed.

02:31 Then I saw, oh, hey,

02:33 cool, Randall's doing that.

02:35 Oh, great timing then.

02:37 Yeah, Kelsey is basically a

02:40 AI search intelligence tool

02:41 to help companies get mentioned in AI.

02:44 So one thing that I've really noticed,

02:47 like a gap that I've been seeing

02:48 is a lot of these tools are just focused

02:49 and obsessing on tracking,

02:51 which is not very useful in AI search

02:54 because it changes a lot.

02:57 You can run the same prompt

02:59 for three different people,

03:00 get three different answers.

03:01 You can run it

03:02 yourself, get different answers.

03:04 And with each model, it

03:05 has different answers.

03:06 So of course we do the typical tracking,

03:08 we're running it across multiple models,

03:11 showing you the actual answer.

03:12 But the bigger opportunity that I see

03:15 is because this is still a black box.

03:18 Like we don't know, we can't say,

03:20 "Oh, this is exactly how you rank,

03:21 "this is what you can do."

03:22 It's not clear like SEO

03:24 quite yet, it's a black box.

03:27 What I'm trying to do is

03:28 basically reverse engineer

03:30 how these companies get cited.

03:32 So instead of, you know, it's going like,

03:34 "Oh, here's your AI visibility score,"

03:36 which I hate, like I hate that so much

03:38 when people say AI visibility score.

03:41 Instead of saying

03:42 that, we're focusing more

03:43 on how you can reverse engineer

03:45 the citations a company has.

03:46 So looking at them,

03:47 seeing what they're doing,

03:48 that's showing up

03:49 repeatedly in the model,

03:51 looking for patterns in that

03:52 and then creating content based on that.

03:54 And that's what I'm doing in a nutshell.

03:57 All right, that's a lot to dig into.

04:01 I mean, in the course

04:03 of the conversation,

04:04 which I don't have like a

04:07 list of questions prepared,

04:09 I kind of just want to talk about this.

04:10 I'm sure I'll ask about

04:12 secret sauce like things.

04:13 Just, you know, tell me,

04:15 well, that's, you know,

04:16 you have to pay for that.

04:19 I kind of want to

04:19 frame why it matters to me

04:22 and why I'm interested because...

04:24 Yeah, I'd love to hear that.

04:26 Was, I had the super

04:28 provocative conversation

04:29 a few months ago with an

04:31 executive who was in charge

04:34 of a DevRel function

04:35 at a well-known company.

04:37 And that person said, well,

04:38 the whole DevRel playbook

04:39 is dead because of AI.

04:41 AI just kills it all.

04:43 And we should just all be

04:44 vibe coding applications now.

04:46 Oh, sure.

04:47 I reacted negatively as one might.

04:50 Yeah, but I thought,

04:51 okay, you know, am I,

04:54 what about my own program?

04:55 Am I really reacting enough to AI?

04:57 Am I just still kind

04:58 of doing the old stuff?

05:00 And in that kind of thought process

05:04 of thinking through all that,

05:06 I got to this problem and

05:08 where my own thinking went

05:11 was SupaBase, oddly,

05:13 Among the hosted

05:14 Postgresses of the world,

05:15 I don't think this is true anymore.

05:17 We're recording this in early July, 2026,

05:19 but for a long time, if

05:21 you asked Claude or ChatGPT

05:26 about something that might have needed

05:28 a relational database,

05:29 it would say SupaBase.

05:31 Okay, it wouldn't say

05:32 RDS, it wouldn't say Nile,

05:34 it wouldn't say Neon, it

05:36 would say PlanetScale,

05:38 like there's options

05:39 but it would always say SupaBase

05:40 I don't know if these guys

05:41 did, but they nailed it, right?

05:42 Yes.

05:43 Part of, as I see what DevRel does is

05:45 the job is to create awareness

05:48 around a particular technology

05:50 that can solve a

05:51 particular problem for a developer.

05:52 And create awareness,

05:54 you have to help

05:55 people understand the thing

05:56 and give them resources

05:58 to play around with it

05:58 and help them build things with it.

06:00 And that's the whole developer's journey.

06:03 And that hasn't all

06:04 been taken over by AI,

06:06 but AI is in the game.

06:08 So sometimes you're gonna ask a question,

06:11 I kind of want to

06:12 build something like this.

06:13 What should the stack look like?

06:15 Well, there you go.

06:16 Yeah, it's so true SupaBase is...

06:18 They're crushing it

06:20 Yeah, in the SEO era,

06:23 which by the way, 25 years ago,

06:24 that was all a scary block of black box.

06:27 That's what I'm counting on.

06:30 Yes, wizardry and

06:33 unscrupulous consultants

06:35 and the whole thing.

06:38 Being the third or fourth

06:39 result was still valuable.

06:41 And now it just, the stakes are higher.

06:44 Yeah, it could be one

06:46 result, it could be two,

06:48 it could be four, it ranges.

06:49 I feel like you just like one to four.

06:51 And maybe you just need to be the one

06:53 that gets recommended, really,

06:55 you kind of need to be first on the list.

06:58 That's not entirely true,

06:59 many AIs will give you lists

07:00 Yeah, it's very different, very,

07:02 I don't wanna say scary times, just

07:04 changing times maybe,

07:05 because you could be on a

07:06 list of one to five companies

07:09 or you could be just one of two,

07:11 or like you said, not mentioned at all.

07:13 And that's something that all companies

07:15 kind of have to start looking into.

07:16 I mean, for me

07:17 personally, I'm sitting now

07:18 and I'm like, "Give me my

07:19 next book recommendation."

07:21 And I asked ChatGPT, and

07:24 then I go buy that book.

07:25 And unfortunately, one

07:26 problem that we've been having is,

07:28 attribution is that

07:30 companies don't really know

07:31 how much is coming in yet.

07:33 For instance, Transistor and Tally,

07:34 they've talked about

07:35 it a lot in podcasting,

07:37 huge amount of search now from ChatGPT.

07:41 And so they're doing really

07:42 well, but other companies,

07:43 they just can't really track it very well

07:45 because we don't always get that

07:47 reference in the URL.

07:48 So that's another kind

07:49 of interesting problem.

07:51 Yeah.

07:52 Why do you hate the AI visibility scores?

07:54 So that's the thing that

07:55 you're not trying to do.

07:56 Don't talk about it. Yes.

07:58 Oh, I hate it so much. It's fake!

08:00 It's completely fake.

08:02 What prompts are they testing?

08:04 What are they using personas?

08:06 What are they doing exactly

08:08 to come up with the score?

08:09 Companies are using

08:10 this, and it's all fake.

08:12 They're making it up.

08:13 What is it based on?

08:14 Are they, "Oh, here's

08:15 your exact set of prompts,

08:17 and here's your score out of them."

08:19 I hate it.

08:20 I hate it so much.

08:21 We're a little more

08:21 focused on the mention score.

08:23 So a set of prompts your company has,

08:26 we generate those for you

08:28 based on your company's website.

08:30 Obviously, you can

08:30 tweak them and add more.

08:32 And then we have

08:32 different layers for each prompt.

08:34 So this is a little

08:34 bit of a secret sauce,

08:35 but for instance, discovery prompts,

08:38 SupaBase, even Vercel,

08:41 getting really early in

08:43 the process of discovery

08:44 for a customer and being

08:46 mentioned those early prompts,

08:48 discovery prompts where it's at.

08:50 And then later on, when they

08:51 start asking for competitors,

08:52 they already know you

08:53 because they were like,

08:54 "Oh, we just got

08:55 referenced this whole blog post

08:56 on why we should be using Vercel"

08:59 Why we should be using all these

09:00 different products.

09:02 But long story short, I

09:03 hate the made-up score.

09:05 What is it?

09:06 Nobody can explain what it is.

09:07 Everyone has their own definition for it.

09:09 It's just kind of,

09:10 I really think it is something the models

09:13 have made up themselves and now

09:14 everyone's using it.

09:15 So I hate it, but it

09:16 does show up high in search.

09:18 So we still have to kind of rank for it.

09:20 So I, you know, SEO does not go away.

09:23 Yeah, it doesn't go away.

09:25 It's a little side quest.

09:26 We don't need to go on right now,

09:28 but it's just an

09:29 illustration of how AI is additive.

09:31 Like you don't get to

09:33 stop optimizing for keywords

09:35 in Google freaking search, right?

09:36 Yeah, absolutely.

09:38 We're still doing it.

09:39 And now you're providing

09:41 this tool that helps you do

09:43 the equivalent for AI.

09:46 We don't have it agreed upon term yet.

09:47 I think in my blog post,

09:50 I called it GEO

09:52 generative engine optimization

09:53 and even sort of parenthetically said,

09:55 I don't know if that's gonna stick.

09:56 Nobody knows this list.

09:57 AEO, GIO, GEO, nobody knows.

10:00 We'll let you know in five years.

10:01 Yeah.

10:04 Someone's gonna do great

10:05 with some domain names,

10:06 but we don't know who yet.

10:08 That's right.

10:08 So, okay, that makes sense.

10:12 In other words, the AI

10:14 visibility score is just opaque

10:15 and not, nobody can

10:19 really get a robust account

10:20 of what's going on there.

10:22 Yeah, it's, you know,

10:23 it's just still a black box.

10:24 We don't know what's going on.

10:25 Everyone is guessing,

10:26 which is why we're so focused

10:27 on seeing what

10:29 competitors are actually doing

10:31 and, you know, taking

10:32 inspiration from that

10:34 and working from there

10:35 instead of just making up,

10:36 oh, you're showing this

10:37 prompt, you don't show this prompt.

10:38 We're like, here is actually this list

10:40 that every competitor is mentioned in,

10:42 which is in our source explorer feature,

10:46 and actually showing them that list

10:48 and showing them what

10:48 ranks up the highest

10:50 and what kind of type of content is

10:51 and then generating

10:52 content briefs from that.

10:54 So, I guess, yeah, walk

10:56 us through the process.

10:57 And I wanna say first,

11:00 this isn't meant to be like a,

11:02 just an ad explicitly for

11:04 this service that you've created

11:06 and you are selling to

11:07 people who find it valuable.

11:09 However, we like it

11:10 when people make services

11:11 and sell them to people

11:13 who find them valuable.

11:14 We love building, yeah, we love building.

11:16 It's all a good thing

11:17 And the plus side is this is also new

11:21 and uncertain and unknown that if you

11:24 just walk us through

11:25 like what's your recommended process

11:27 for using Kelsey,

11:28 honestly, I just think that's

11:29 informative to the audience.

11:30 I'm gonna learn something, so.

11:31 Yeah, no, I thought that was—

11:32 How should it go?

11:33 (laughing) You mentioned discovery

11:35 prompts and a few things

11:36 and report and just walk

11:38 us through the whole thing.

11:39 Yeah, yeah, so basically you sign up,

11:42 you add your company website,

11:44 and we start pulling

11:45 prompts from all of your content

11:47 of what your potential

11:48 customers might be searching for.

11:50 And then we build,

11:51 what I personally think

11:53 is the future is topics

11:54 instead of individual prompts.

11:56 So for instance, well, I

11:59 should specifically go on

12:00 to users that are

12:01 using the tool right now,

12:02 but we have—

12:03 Right, some company.

12:04 Yeah, for instance, okay, one company,

12:07 they are selling to other founders,

12:11 smaller B2B solo founders,

12:14 and we would start by

12:16 having them enter their website

12:18 and then we start

12:18 pulling topics for them.

12:20 So for instance, a B2B

12:22 SaaS boilerplate, for example,

12:24 we're pulling what is the

12:25 best boilerplate to use?

12:27 What is the most

12:28 secure boilerplate to use?

12:29 What is the best React

12:31 and Node boilerplate to use?

12:33 What is the best Python boilerplate?

12:35 And then we're generating overall topics,

12:38 and then from there we drill

12:39 down into individual queries

12:41 that go onto the whole customer journey.

12:43 So starting with discovery

12:45 all the way to the customer

12:47 saying, okay, well, tell

12:47 me about these competitors.

12:49 And I feel like the key

12:51 there is going really deep

12:52 into the kind of longer, I

12:54 don't wanna use SEO words here,

12:56 but like the longer tail prompts

12:58 where someone is just starting out

13:00 and they are trying to do

13:00 discovery and trying to build,

13:02 and there's a lot of opportunity there

13:04 to kind of win those early prompts,

13:07 and then in the future

13:08 try to get the bigger ones

13:10 because you can't really compete

13:11 with the big companies right away,

13:14 but you can't compete

13:15 with like the early on

13:16 in the discovery,

13:16 which is what we're

13:17 trying to help people do.

13:19 Those prompts, you said you

13:21 analyze text on the website

13:23 and you're inferring the kinds of prompts

13:25 that would generate that as a result.

13:28 Yes, so we're pulling

13:29 all our website data

13:30 based on what they're doing,

13:32 what people might be looking for,

13:34 for instance, like what is the best React

13:38 and Node boilerplate I can use?

13:40 And from there, that's maybe one topic,

13:43 and then it's gonna be like, okay,

13:44 should I be using ShipFast?

13:45 Should I be using Gravity?

13:47 You know, what are my

13:48 security considerations?

13:50 And there's a lot of

13:51 prompts that go into that,

13:52 and then for each topic, we have a

13:55 million prompts in that.

13:56 So you can search

13:57 through all this giant layer

13:59 of discovery prompts all the

14:01 way to buyer intent prompts.

14:03 We run those for each topic,

14:05 so you have kind of a

14:06 full comprehensive view,

14:08 and then from there, your data's all

14:09 populated in Kelsey.

14:11 And we kind of start

14:12 getting a lot of information.

14:13 We can see the top competitors.

14:14 We can see the top-sided page

14:16 that's being shown over and over.

14:19 And, well, I'm really excited about it,

14:22 but we just released a new

14:23 feature about content hubs.

14:25 So those are really big that we've found.

14:27 Everyone's talking about

14:28 this right now in AEO, GEO,

14:31 whatever you're gonna call it, AI search.

14:32 What's a content hub?

14:34 Yeah, so a lot of companies

14:36 that are doing these are ranking higher,

14:38 and it's basically they're covering

14:39 the entire journey to one prompt.

14:42 So all the way from the

14:43 discovery to the buyer intent,

14:45 they're covering every

14:46 question the buyer might have.

14:48 And so that's how the LLM is like,

14:50 oh, wow, this person is

14:51 extremely knowledgeable

14:52 because they can answer everything.

14:54 And, you know, HRF is

14:56 talking about this in their course,

14:58 you know, which is smaller,

15:00 like in Semrush—Semrush as well

15:02 Everyone's talking about

15:03 content hubs right now.

15:04 So that's why I'm

15:04 excited we're generating

15:06 basically large content briefs

15:08 for building your own content hub.

15:10 but I wanna back up and

15:12 make sure I understand

15:13 the prompt thing.

15:14 So your tool analyzes

15:16 the text on the website

15:18 and sort of imagines for, you know,

15:22 past and AI to say what

15:23 prompt would be answered by this?

15:27 And then you actually run that prompt

15:30 and see what the results are.

15:31 Yes.

15:32 You know, is it your

15:33 website or is it these other 50?

15:35 That's basically that process, right?

15:37 Yeah, very comprehensive

15:39 because we were running

15:40 it across multiple models,

15:41 sometimes multiple queries,

15:43 even at the same question.

15:44 And then we layer on that personas.

15:47 So for instance, for me,

15:49 when I'm asking, you know,

15:50 what form builders should I use,

15:52 what should I use for forms?

15:53 I'm gonna see like Tally,

15:55 because you know,

15:56 they're huge in a dehacking,

15:57 but someone else might see Google Forms.

16:00 So, you know, with each prompt, yeah.

16:02 So that's tricky because you can think

16:04 you're doing really well.

16:05 Like a big company is

16:06 gonna go search themselves

16:07 and think, oh, wow,

16:08 we're really crushing it.

16:09 But to another person,

16:11 they're gonna get

16:11 completely different, you know,

16:13 set of answers for each prompt.

16:16 And that is because of the memory

16:18 that the models have recently.

16:21 The broader context across

16:25 chats that we've developed.

16:27 Yes, so that is why we

16:28 have the personas feature

16:29 specifically to help kind

16:30 of indie founders like that

16:32 as well, even large companies,

16:33 because for each query that we have,

16:35 you can add a persona onto that

16:37 and see kind of where

16:38 you show up based on that.

16:39 So what then, again, I'm

16:40 just trying to make the analogy

16:41 to the early days of SEO.

16:43 And I was anything

16:44 but an expert back then.

16:45 Like I was reading,

16:47 Matt Cutts was the big guy.

16:48 He worked for Google and he

16:49 was sort of teaching everybody

16:51 what SEO was and

16:52 separating hype from reality

16:55 and the disreputable folks

16:58 out there in the margins.

17:01 And I remember, this is

17:03 kind of a, I was there,

17:05 Gandalf, I was there 3000

17:06 years ago kind of story,

17:07 but in 2005.

17:10 Not long ago, yeah.

17:13 And it was things like, okay, well,

17:14 your markup should be semantic,

17:16 which was another new

17:17 idea that, you know,

17:18 and don't put everything in tables

17:21 and just write good content.

17:24 And the kind of rules that emerged as,

17:28 the things you should

17:29 do with your content.

17:32 Thankfully, good

17:32 content stuck around at least.

17:35 Yeah, make good content and represent it

17:38 in a clean way on the page.

17:41 Are there any emerging practices?

17:43 Like clearly somebody's

17:44 gonna need the kind of tools

17:46 that you're creating to

17:47 give them that visibility

17:49 and all the layers of,

17:51 like the persona layer

17:52 that you're doing,

17:52 that makes perfect sense.

17:54 And the ranking against

17:55 competitors and all that.

17:56 You need a tool that's gonna

17:58 do that in an automated way.

18:00 But then what do you do?

18:01 So my kind of, my heart

18:02 here is how are DevRel teams

18:06 and DevRel

18:06 professionals responding to AI?

18:09 How should we?

18:10 And it's a broad theme

18:11 that I try to articulate is,

18:13 you're creating content for people

18:15 and you're creating

18:16 content to teach the models.

18:18 And you don't teach the

18:18 models if they don't see it.

18:20 What do you do?

18:23 I mean, suppose you're not ranking

18:24 exactly where you want,

18:25 you wanna move up, is there

18:26 an emerging set of practices

18:28 for being better?

18:29 Yeah, well, there's two aspects of that.

18:31 Like now people have

18:32 to focus on showing up

18:33 in like the training data.

18:35 So they have to be,

18:36 still doing their SEO,

18:37 still doing Reddit, showing

18:38 up on Quora even, YouTube.

18:40 And then they also have

18:41 to be showing up in the,

18:43 the current recent search.

18:45 So the recent search that's going on,

18:47 is trying to pull in the freshest data.

18:48 So there's kind of two

18:49 aspects they have to be thinking

18:50 about and then it is a lot, honestly.

18:53 It's like SEO just plus, SEO plus.

18:56 Hopefully that's not the

18:57 term that people choose

18:58 because that would be

18:58 terrible, but it's, you know,

19:01 they have to be thinking

19:02 about being active on Reddit

19:04 without getting banned, which is hard.

19:06 They have to be doing

19:07 YouTube videos, you know,

19:08 that still add value, websites,

19:11 it's still add value

19:11 to people, but you know,

19:12 the agent can read and

19:15 still creating good content

19:16 that everyone wants to be

19:18 looking at, but is readable.

19:20 It is a big thing that we're focused on.

19:22 One of the big things, a really easy

19:24 thing to do immediately,

19:25 I think for your

19:25 website is for every header

19:27 should be answering a

19:28 question in a really quick way

19:31 that AI can take it like

19:33 a quick takeaway for AI.

19:35 So you should be

19:35 mentioning your company's name

19:37 in a sentence and then

19:38 saying exactly what you do.

19:40 So you're almost like

19:41 you're answering a question

19:42 for AI to read at a glance,

19:44 which is why FAQs kind of help.

19:47 People show up, we have

19:48 FAQs on quite a lot of pages.

19:50 We're already showing up at AI search,

19:51 even though, you know,

19:52 we've recently launched.

19:54 So that's a big one,

19:55 but it's a broad topic,

19:57 which is, you know, why

19:57 we're trying to compete,

19:59 reaching out to

20:00 people that have listicles

20:01 and trying to get, you know,

20:03 published in those as well.

20:04 That's a big one, not just, you know,

20:06 publishing your own listicles.

20:08 Comparison pages are huge

20:11 because they will actually go read those.

20:13 The AI model will read them.

20:14 And then even pricing pages are big

20:17 because they will create a

20:18 comparison from that as well.

20:19 So it's like, it's a lot

20:20 of layers there, I guess.

20:22 Yeah, yeah, but the thing

20:26 that struck me as a header

20:28 with a relatively complete,

20:30 and I'm not sure if I'm making this up

20:33 or if I can say it's like not too long,

20:35 a good concise answer under the header.

20:38 Yeah, so when you're

20:39 creating a long blog post,

20:40 you can't really be,

20:42 you know, going into it.

20:43 I mean, yes, please

20:44 go into a ton of detail

20:45 and make great content,

20:46 but you need to give the

20:47 AI like readable chunks

20:49 that it's gonna take.

20:50 You can't get to your point

20:52 and, you know, one sentence

20:53 and then 20 sentences later be like,

20:55 oh, also our company does this.

20:57 You have to have

20:57 really concise statements

20:59 for the AI to pull and understand.

21:02 Yeah, and that sounds like that comports

21:06 with the chunking

21:07 process on the ingest pipeline

21:09 of pre-training where

21:11 it's, I imagine it's smart

21:13 with respect to the

21:14 structure of the document.

21:15 You know, here's a heading

21:16 and here's some stuff

21:17 underneath the heading.

21:18 Let's make that a chunk.

21:19 Yeah, and I mean, I

21:20 know Google is saying,

21:21 oh, just create good content.

21:23 We're gonna punish

21:24 everyone doing all these tricks,

21:26 but one, they've been

21:27 saying that for a long time,

21:28 and two, I don't

21:28 necessarily trust everything.

21:30 Google says sometimes.

21:32 They just want us to

21:33 create a ton of content,

21:36 which is great, but we're

21:38 not, you know, people are not,

21:39 we've noticed companies are not getting

21:41 the same amount of clicks anymore

21:43 because of AI overviews.

21:44 So, you know, something needs to change.

21:47 Traffic is falling, yeah.

21:49 On the last few years.

21:51 Yes.

21:52 With the numbers I

21:53 personally see, and that hurts,

21:55 but a lot of that's the Google summary.

21:57 So thank you, folks.

21:59 Yeah.

22:01 And some of it is people

22:04 not Googling, but asking AI.

22:07 Yeah.

22:07 Which you did, which is

22:08 why Kelsey is indispensable.

22:12 Yeah, well, I hope so,

22:13 but that is one big thing

22:15 is that nobody is getting as

22:17 much traffic as they used to,

22:19 but they could be getting, in some ways,

22:22 much higher buyer intent traffic

22:24 because I think Hrefs

22:25 recently mentioned this.

22:26 They get like 23 times higher for them,

22:29 someone who comes in through an LLM

22:32 on if they're gonna

22:33 buy the product or not.

22:34 And I think, gosh, Transistor said

22:36 they had 50% of their

22:38 trials last year come in from AI,

22:42 50%!

22:43 So, yeah, I'm taking on all

22:45 this data myself personally.

22:47 I'm very excited about it.

22:48 And all this kind of leads to,

22:52 let's sort of hint at this,

22:53 AI changing the way we write.

22:55 And I haven't dug into how SEO,

23:00 basically Google changed how we write.

23:03 I'm sure that's been

23:04 analyzed and remarked upon

23:06 SEO optimized content

23:08 isn't just what one might write.

23:12 It's a different thing

23:13 that's had an effect on prose.

23:16 Again, I don't, I haven't

23:19 read any analyses on that,

23:20 but clearly that's a thing.

23:22 And here's another layer

23:24 where writing for the models

23:26 is going to change the

23:28 kind of prose we generate.

23:30 And you can love that

23:33 or you can hate that.

23:34 Yeah.

23:35 My own writing doesn't

23:39 tend to be overly simple.

23:41 You know, it's not

23:43 like a Victorian novel.

23:45 (Kelsey laughs)

23:47 The word and desperately in love

23:51 with the sound of their own voice.

23:54 Most of them are not my

23:58 favorite for that reason.

24:00 And then you've got

24:01 kind of a 20th century

24:04 benchmark of Strunk and White

24:06 and this very minimalist

24:08 Just eliminate anything that's not

24:10 absolutely necessary.

24:12 And if you hold to that,

24:14 you probably don't mind

24:15 the constraints of SEO

24:16 and AI optimization,

24:20 but it's kind of a question

24:21 you have to just come to terms

24:23 with if you're writing for

24:24 yourself, do I want to be me?

24:28 Do I want to

24:28 compromise that a little bit?

24:30 You know, it's an interesting thing.

24:32 I mean, I think it's

24:32 almost like, you know,

24:33 SEO in 2005.

24:35 You should write your content,

24:37 you should write good content

24:39 and then just tweak it a

24:39 little like, you know, in the past,

24:41 maybe we're adding

24:41 little keywords from Hrefs

24:44 and now we might be going to Kelsey

24:46 and editing a little bit

24:47 of our content brief.

24:49 Yeah, yeah, I think

24:51 that's a reality that,

24:54 and as a writer, again,

24:55 it depends on your goals.

24:56 I've always remarked

24:57 that like corporate comms,

25:00 professionals do things

25:01 like write press releases.

25:04 You know, there's a certain voice

25:05 that they're trained to write in

25:06 that I personally am incapable.

25:08 Like I would be a more capable writer

25:10 if I could switch gears into that.

25:12 Oh yeah, my blog was for me.

25:14 Exactly, exactly.

25:15 My writing is me being me

25:19 and wanting to be me authentically.

25:21 Yes.

25:22 This gets to like, you know,

25:23 anonymous employee surveys,

25:25 the quarterly pulse survey.

25:27 And if I get past about

25:28 five words in a comment

25:30 and you know me, we're all gonna know.

25:35 I think I have a fairly distinctive voice.

25:40 I feel like I don't wanna give that up.

25:42 And I think the catch here is that

25:43 you don't have to give

25:45 that up, you get to be you.

25:46 And part of the reason why

25:48 you're writing is to be read.

25:51 Most of us want to engage with readers.

25:53 Yeah.

25:53 You want that well?

25:56 Yeah, there's definitely, yeah,

25:57 there's something to it now.

25:59 It's like, it has to change in a way.

26:02 But I mean, for me, I'm

26:03 still like, my personal blog,

26:04 I'm kind of relaunching it right now.

26:06 I'm still writing for

26:07 my audience to help them.

26:09 I'm not focusing on AI.

26:11 I don't, I hate to say it,

26:12 but I'm focusing on my audience first.

26:15 And then I might tweak a few headers.

26:17 I might add an FAQ,

26:20 but I'm not going to

26:20 completely change who I am

26:21 because, you know, I

26:23 still want to add value

26:24 in my own unique way.

26:25 I don't want to be sitting

26:26 there trying to, you know,

26:27 obsess about what AI

26:29 model is going to read.

26:30 Because, you know, as

26:31 much as we are losing clicks,

26:33 we're just unfortunate in losing people

26:34 going to that website.

26:35 I still think there

26:36 is something, you know,

26:37 to adding valuable

26:39 content, but at the same time,

26:40 just tweaking it a little bit.

26:42 So it shows up.

26:44 Yeah.

26:44 Can be read.

26:46 Yeah.

26:47 Does Kelsey give that kind of guidelines?

26:51 I mean, it gives you

26:52 analysis of where you stand,

26:53 but does it give you advice?

26:55 You'll scan your website and say,

26:56 "Wow, this sucks. You

26:57 should blah, blah, blah."

26:58 Yeah.

26:59 We do have a few, like kind

27:01 of, I'd say, like setup steps.

27:03 So in our kind of recommendations,

27:05 going through like what

27:07 you can do in small ways

27:08 to kind of help your website show up.

27:09 Like a lot of websites

27:11 with Cloudflare Automatically

27:12 are blocking robots

27:14 from reading their site,

27:15 and people don't know that.

27:16 So checks like that really

27:18 important, you know, show up.

27:20 And then on the other layer of that,

27:23 we go through in, you

27:24 know, the actual prompts.

27:27 And once we've created a

27:28 content brief from that,

27:30 you can go through and see,

27:31 "Okay, these are the

27:31 headers I should have.

27:32 These are the exact

27:33 questions I should be answering.

27:35 These are the FAQs that

27:37 will help kind of the model

27:38 read that.

27:39 And these are all based on

27:40 all the data we're pulling in,

27:42 you know, per website prompt.

27:44 So all the competitors, all the lists,

27:46 and kind of seeing who's

27:47 ranking and then going from there.

27:49 So kind of two aspects of like, you know,

27:51 the overall setup, what

27:52 pages you should create.

27:54 Should you be doing

27:54 competitor pages, comparison pages,

27:57 alternative pages, listicles,

28:00 starting from there,

28:01 and then actually drilling down

28:02 and creating those content briefs,

28:04 where it's like, you

28:05 should be creating this header

28:07 and you should answer this question.

28:08 And then, you know, still go forward

28:10 and create great content,

28:11 be individual, you know,

28:12 be an individual still, have

28:14 your company voice, you know.

28:16 But just going through and making sure

28:18 you fill those layers,

28:19 which is a big reason we don't,

28:20 which another thing I hate

28:22 and I think is everyone's gonna

28:23 get punished for this in like a year.

28:24 Sorry, am I just talking

28:25 about like everything I hate

28:26 in this podcast?

28:28 (laughing)

28:29 Be opinionated!

28:30 I love it

28:31 Well, one thing I hate is

28:33 all those companies right now

28:34 that are trying to

28:35 generate giant blog posts

28:38 to rank in AEO.

28:40 I just, I hate that

28:41 because they're giving you

28:43 the full blog posts and they're like,

28:45 here, just publish this.

28:46 No, everyone has a unique voice.

28:49 Every company has their own brand voice

28:51 and they have their own

28:52 way, their own, you know,

28:53 own knowledge of adding value.

28:55 And we don't do that.

28:57 We do a content brief

28:58 where you know headers,

28:59 you should add sections, you should add,

29:01 if you need a

29:01 comparison table, if you should,

29:04 you know what links you might need.

29:05 I don't believe in

29:06 giving the full blog posts

29:07 because I think it is

29:09 important to still answer

29:10 in your own voice because

29:11 I think all those companies

29:12 are gonna get punished and

29:13 yeah, it's gonna, you know,

29:14 it's gonna be a massive drop in rankings

29:17 and they're gonna be all running around

29:18 chickens with their heads cut off.

29:20 Couldn't agree more, but

29:21 from a DevRel perspective,

29:22 I think the fact that AI

29:25 brings the marginal cost

29:26 of creating content a lot closer to zero

29:30 and there's all this

29:30 slop and that's like,

29:32 you're making people sad that it's,

29:34 there's all this stuff to sort through

29:35 and not the people that want it.

29:37 The value of being a

29:38 person who is trusted

29:40 and, you know, a influencer type or

29:43 developer advocate type

29:44 or something like that.

29:44 You don't have a

29:46 relationship with a community

29:47 or your audience like

29:49 you do with your family

29:50 or your friends, it's

29:51 not a symmetrical thing,

29:53 but dang it, it is a relationship.

29:56 And the trust that they

29:58 can build in you as a person,

30:01 all of a sudden means a lot more.

30:03 Yeah, that's--

30:04 So, hanging on to that

30:05 becomes much more important.

30:07 Oh, I love you said that.

30:07 I don't think you're saying anything

30:08 about AI optimization

30:11 that would suggest that we

30:12 should do away with that.

30:13 Yeah, no, I love that you

30:14 said that's all we have anymore,

30:16 right?

30:17 All we have now is trust

30:19 that someone is a real person

30:21 recommending this product.

30:22 That's all there is left.

30:24 Distribution, trust, that's it.

30:27 I really believe that.

30:29 Yeah, well put.

30:30 Hey, how can folks follow

30:32 you, find out more about Kelsey?

30:34 You would give us links,

30:36 I'll make sure they're in there.

30:37 Yeah, yeah, company

30:39 website, trykelsey.com

30:41 and my Twitter, I'm

30:42 always on there talking

30:43 about kind of my founder's journey,

30:44 which is @randallkenna.

30:47 Well, my guest today

30:48 has been Randall Kenna

30:50 Randall, thanks for

30:51 being part of the show.

30:51 Thank you so much for having me.