I recently recorded a conversation with Jeremy Rivera for his Unscripted SEO podcast, and this is the full transcript.
My position going in: GEO and AEO are extensions of good SEO, and the biggest new factor is what the rest of the web says about you. Jeremy backed that with data from Ben Wills’ OppAlerts study on topically aligned backlink profiles. We also covered why links matter more now than five years ago, how I map blog posts to service pages (and why that meant cutting 60% of one client’s blog), and why the four-blogs-and-two-links retainer no longer works in competitive niches.
The second half is less optimistic. We talk about LLM output polluted by a low-quality web, a fractured SEO community, and AI drafts that need so much editing that the time savings are smaller than they look. That is why I’m building Assertio, which writes, rewrites, and scores content at the claim level so it actually says something instead of producing AI slop.
The key takeaways are at the bottom if you want the short version, and the video is here:
The Transcript
JEREMY RIVERA
Hello, I’m Jeremy Rivera, your Unscripted SEO podcast host.
I’m here with Lum Kamishi, who’s working with Assertio, and that’s Assertio.ai.
And I’m going to ask you to tell me who you are.
LUM KAMISHI
Hey, thanks for the opportunity.
Well, I’m Lum.
I’m an SEO guy based in Kosovo, that’s in Europe, somewhere in the Balkans.
I’ve been doing SEO for about 11 plus years.
I initially started with a friend of mine.
He was running a really small agency with a couple of clients.
That’s how I got in.
I was studying math at the time.
That’s when I realized that math’s not for me really.
So SEO came to the rescue, or maybe not the rescue.
Maybe that’s the downside of me.
Right now I’m a contractor for an agency based in Scotland, where I run a few campaigns and clients, and I’m also building and developing Assertio.ai, which is my own project.
It started off from a few ideas that I had throughout my whole career in SEO, and of course I’m happy to share that with you here tonight, or today.
It’s evening here, so yeah.
BEST QUOTES
“GEO is basically what the web is saying about you, right?”
— Lum Kamishi
“It’s not that links now still matter, but in fact they matter more than five years ago.”
— Lum Kamishi
“I don’t care about budgets, I care about results.”
— Lum Kamishi
GEO is SEO plus digital PR
JEREMY RIVERA
Awesome.
The time changes, so it’s a tricky one.
Tell me what it is that you’re seeing right now that matters for visibility and AI.
I think of this equation as: there’s fundamental core pieces of SEO, and they’ve all been enveloped over the top with additional extra stuff.
So everything that we have done for SEO still matters, but then there’s a bunch of extra stuff that’s also pushing the balance here and there.
And some people have theories as to what that might be.
All sorts, back to some of the research that I’ve seen come out recently from OppAlerts and Ben Wills and others that helps clarify a few of those things.
But if you’re building something, that means you have a pretty good supposition, and as you’ve been building it, I’m sure that you’ve been challenged on the output that you’ve been able to get out of your systems.
So what’s your foundation for AI optimization?
LUM KAMISHI
Well, I’m more of a conservative guy.
So I really believe that, in the bottom line, GEO or AEO or these acronyms is basically just an extension of good SEO.
And by SEO I don’t just mean fixing technical stuff and sitemaps and everything like that, but a cluster of a lot of techniques and a lot of stuff that we used to do to bring the best results.
That includes content, external citations, technical foundations of course, that’s your baseline.
And I’ve been seeing a lot of SEO professionals suddenly selling GEO and AEO, because that’s what clients are looking for, right?
They hear a word, they want to get it.
Just to make one small remark: I worked in-house at a company for about eight years, so no client work, right?
I was one hundred percent focused there.
Then in 2025 I started taking on clients, and that’s when I realized that these people really want to get a GEO service sold to them, right?
That’s what they’ve been hearing online, and that’s what they want for their company.
So if I go around and tell them that’s just good SEO with some extra steps and a holistic approach to SEO, they would not really buy it.
“Yeah, I want GEO, I want to show up in LLMs. I want those fancy techniques that everyone’s talking about.”
But in my opinion, from what I’ve seen so far and the results that to an extent I’ve been able to reproduce, GEO is basically what the web is saying about you, right?
So the primary factor in LLM and AI rankings are citations, backlinks.
And by citations I do not mean those local directory citations, but citations and mentions from niche-relevant publications with relevant content pointing to your site.
Not necessarily a link.
A link would be really helpful.
But a network of publications that speak about your product, your site, your services and whatever.
Reddit used to be really big on this, but last month or so there was this really big crackdown from ChatGPT.
It suddenly stopped using Reddit as its main source, or its prevalence in AI answers.
So I would go like: GEO is SEO plus digital PR.
And digital PR is something that we, at the company where I used to work, first discovered in 2017-18.
Since I’ve been familiar with that, it was really not something new to me when I started learning about these GEO concepts and notions like network theory and stuff like this.
So basically just SEO, classic SEO, with digital PR and some forums maybe, Quora, Reddit.
GEO = SEO + digital PR: the baseline plus what the web says about you.
JEREMY RIVERA
I think you’re right.
LUM KAMISHI
Yeah. Thanks.
The evidence: niche-relevant links and AI visibility
JEREMY RIVERA
And I can give you some evidence.
I can give you some proof.
So I interviewed Ben Wills individually, formerly from Ontolo, and then he started his own thing, OppAlerts.
It’s another AI visibility platform attempt.
But he did a huge, massive data study to try to identify: all right, I’m going to go across multiple different models and put a ridiculous number of queries in a ridiculous amount of times.
And he came back with a ranking factors study of these various tests that he tried to look at: relevance of the home page, overall embedding relevance of the entire site.
He also used vector embeddings to look at backlink profiles of sites, as well as branded linked mentions and unbranded citations.
And he found, not saying that there weren’t statistical signals for on-page relevance or things like that, but he said that there were twenty thousand cases where it was a positive match and zero cases where it was a negative match for your backlink profile being from pages that were in the same vector embedding category.
So if you’re in the digital marketing space, your links all come from sites in the digital marketing space.
If you’re in the home improvement space, your backlinks come from sites that are in the same vector embedding as being about home improvement.
LUM KAMISHI
In a certain location, yeah.
JEREMY RIVERA
So yeah, I’m going to add the link to the show notes if anybody wants to dig into that.
Certainly anybody that wants to challenge it from a data perspective.
I spent some time digesting it and understanding it.
It makes sense to me how he broke down his approach: your individual page gets turned into text, from text into numbers, and that vector embedding is actually fairly similar across multiple different models, and so you still end up with that same topical map.
So it makes sense to me that backlinks are a signal.
I think they’re used more naively or more generally than in Google, because Google has had to add layer and layer and year and year of defense against very large and very powerful and very effective black hatters who find the weak underbelly.
And Google, for a long time, they played whack-a-mole, and they said, “Well, we’ll punish you out of existence, we’ll use Penguin to punish you for anchor text abuse.”
So we know that the link graph in that side of things is extremely helpful for retrieval engines and for relevance.
There was a moment there where it’s like, backlinks don’t matter anymore.
And I never really bought that.
There was too much alignment between popular and big and authoritative brands and sites that would show up in these citations.
If it was truly just what was relevant and not based off of authority signals, then the citations would be wild.
They’d be going all over the deep side of the internet and pulling random things from tiny little sites all over the place.
But that’s not what we see.
We see citations from relatively normal, authoritative sites with decent topical backlink structures.
LUM KAMISHI
Exactly.
Actually, my first work in SEO was a Penguin recovery program that I did for a client.
So those backlinks, we have to mention this, those kind of backlinks that used to work in 2014, of course they do not work anymore and will get you penalized in Google.
And LLMs are basically way more primitive than Google when it comes to ranking.
Ranking in quotation marks, or selecting or recommending, because Google, as you said, has been playing whack-a-mole and has perfected its algorithm against people who did link building for a living.
And I really think that ChatGPT and Perplexity and all these guys are really way behind on that.
So they’re really more prone to being manipulated.
And in fact, it’s not that links now still matter, but in fact they matter more than five years ago, right?
Because five years ago, I remember pre-COVID times, the big thing that we were discussing was BERT and MUM.
We were panicking because now it’s all about content, core updates.
I think that’s the time when they started rolling them out, 2019, 2018.
So content was the big thing, and I really think that with the LLMs now, link building is doing a resurrection.
There’s a resurrection of links, but good links, niche relevant.
That’s links that have similar vector embedding coordinates, like cosine similarities really, really close to the site as a topic and to the network as a whole.
So I call that niche relevance.
I’m a classic guy, so yeah.
Link building now: you pay for the link
JEREMY RIVERA
I wonder what it is that you’re putting on offer?
Because it is hard.
So going back to the concept of link building, doing it to get links from real, authoritative, niche-relevant sites, it’s not easy anymore because everything’s been so spammed out.
We reached a point in the ecosystem where—
LUM KAMISHI
I don’t do it. Yeah.
JEREMY RIVERA
—almost nobody responds to an email, because you used to be able to reliably say the domain’s connected to this email, if I send an email I’ll get a response.
But it’s—
LUM KAMISHI
Yeah, now they just send you the bill, right?
“This is how much it would cost.”
Personally, I don’t do link building anymore.
In fact, I haven’t done it since pre-COVID, but I’ve managed link builders, and I have friends who are good link builders to this day, and all they can say is that they have to pay now for links, right?
So the cost has gone up.
Topical proximity and cannibalization
JEREMY RIVERA
The other half of the equation: one part is getting that visibility, but you also have the other side of, how do you structure relevance?
How do you create robust content creation systems?
What’s your mindset right now about content creation?
LUM KAMISHI
I’m really, really hands-on in content and in technical stuff.
That’s my area of expertise.
I use a lot of approaches based on the needs of a client.
So if a client wants to convert, or if a client wants to build topical authority, I’d use different approaches to each of them.
Of course vector embeddings come into play.
But something I did for a home services site, a home renovation site: I mapped out their blog posts to their core services, right, and calculated something I call a topical proximity, which is the semantic distance between a blog post and a service page.
So if we have a service page, if we have a service offering roof renovations or offering bicycle renovations, I would usually create a mini cluster of two or three blog posts that would assist that service page.
And of course, keep in mind cannibalization.
Not in the classic sense, where you have keywords competing against each other, but in my case I used topics competing against each other.
So I wouldn’t want a blog post competing against a service page.
An issue that I constantly see is blog posts ranking for money queries, right?
When in fact a service page should have ranked.
So I try to pay attention to this, and with finesse and everything I try to angle the blog post in a way that it doesn’t compete with a service page, but also doesn’t roam too far away from what that core service page is all about.
So for the money queries, I want the service page to rank and to show up in LLMs.
I use the blog posts as just a helper, or some wheels for the service page.
Topical proximity: blog posts as helper wheels for one service page.
So: topical proximity, cannibalization, avoiding cannibalization.
For this, vector embeddings and cosine similarity is just one layer that I use.
Something that I did is outlining formulas, because as I said, I’m a failed mathematician, so why not do some of that stuff here and fail in SEO as well?
So I like to develop my own formulas to calculate similarities, not just by vector embedding and cosine similarity, but also in terms of larger groups of topics, right?
And yeah, what I’m explaining here cost the client about sixty percent of their blog.
So I killed sixty percent of their blog and added some articles that would also convert but also support their service pages.
The four-blogs-and-two-links retainer is dead
JEREMY RIVERA
I think there’s an underappreciation for the flexibility of the type of content that you can create on your website.
And I think a lot of agencies got stuck in the mindset of, “Hey, I’ve got a retainer and I can do about four blogs a month with paying this writer and taking SEO.”
So we fell into the pattern of, well, I need links, so we’ll get two links per month and we’ll publish four articles per month.
But I think those days of that actually working are long gone.
Would you agree?
LUM KAMISHI
Yeah, I agree completely.
I promised the client at least thirty new blog articles.
And the way that I work, when I work independently, I don’t usually go on promising budget-scope things.
I don’t care about budgets, I care about results.
So if I believe that thirty blog posts would bring you better results, then I would promise thirty blog posts and I would map them out.
If there’s thirty-one, then I would do thirty-one, right?
And I honestly think that this budgeting may be a thing of the past, but in some cases it may still work.
Especially in industries or landscapes where there’s less competition.
Maybe some local niches, some regional markets or stuff like that.
That may work.
But turning to those sites where competition is fierce, niches where competition is fierce, I’ve had my doubts on that.
So really not for this agency-style budgeting, but if it works, it works.
Polluted LLMs and the missing SEO voices
JEREMY RIVERA
I’m curious, what are some of the fringe cases that you’ve seen, or outliers, things where the patterns that we’ve seen just don’t seem to fit?
Let me put it another way.
What in SEO is making you scratch your head and say, what is going on?
What is an active challenge for you in the industry right now?
LUM KAMISHI
Well, everything is a challenge.
What I just described earlier about content creation and building topical maps and content strategies, that’s a challenge on its own, right?
Because even if you turn to LLMs, the results are polluted.
You can try and tell Claude to assist you on a topical map, and the result will be garbage, right?
Because the web is polluted with information of not the highest quality, I may put it that way.
Back in my day… I sound old with this, “back in my day,” three times repeating this.
But back in my day, we had the big guys: Moz used to be really good, Ahrefs used to be really good.
Guys like Rand Fishkin, Cyrus Shepard, Gianluca Fiorelli, all these people.
You knew who to turn to when you had a challenge or something.
And right now you just have to figure it out on your own.
You don’t know who to rely on, what to rely on.
And pretty much maybe this is the challenge: the fact that you can’t rely on anything anymore.
In June I was at a conference in Croatia.
I was really shocked with the doom and gloom, and everything was AI agents automating your workflow, automating this, automating that, agents building, I don’t know, stuff that maybe sells better, but I’m not sure there’s any value to it.
And so yeah, I really miss hearing from good SEOs what they’re really doing in their day to day.
I had the pleasure of meeting Gus Pelogia there.
He was a breath of fresh air, both personally and professionally.
I had a really good time discussing ideas with him.
JEREMY RIVERA
I love Gus.
LUM KAMISHI
Because he works at a job marketplace, and I have my fair share of experience and specialization in marketplaces.
So I had a lot to ask, and he filled me with hope for the industry.
And yeah, maybe there’s still people you can turn to, but I’m still hoping that this whole AI cloud or pollution or smoke or fog will go away, and we’ll be back to, again, fourth time, the old days of SEO.
JEREMY RIVERA
Ha ha ha.
From blog comments to Slack: a fractured SEO community
JEREMY RIVERA
I feel you on that.
As somebody who started in 2007, I grew up in SEO with the original vandals.
We were talking real hardcore SEO on Twitter, throwing down conversations and challenges to each other, before the platform was seen as toxic.
And it was crazy, because you could put up a blog post and people would actually reply with real things, like 800, 1,200, 2,000 words’ worth of content into your comment.
And there was a genuine expectation of the type of dialogue that we now automatically assume happens on social media.
Nah man, the original social media was everybody’s different sites.
Everybody had their sites, every site had a niche community, people would communicate with each other, you’d echo across on social media.
And there’s just been such a hollowing out as the bigger networks and platforms have trained their algorithms to turn that into monetization.
It happened first with Meta, 2010, 2012, around the Cambridge Analytica times.
You’d be shocked that Facebook pages used to be the place for small businesses to engage and network with clients and win business organically.
There was organic search within Facebook.
There isn’t now.
It’s all pay-to-play.
It’s all maximum value derived upstairs into Meta.
And all of it’s hardened on almost all the social media channels.
Reddit went from being a silly place to a silly and expensive place.
The motivation used to be this for SEO: prove your expertise, say what you’re doing, put out unique, relevant stuff, and people would challenge it and there’d be conversations.
But with so many new people entering the industry, the motivation to actually put your homework up on the internet slipped.
And as it entered towards 2020, you would get this class of SEO publishing mostly from SaaS outfits, trying to publish out there to show they were trustworthy so you’d buy their software.
But most folks that were really doing it had moved off of socials.
Maybe it’s in a Slack channel.
Maybe some of the more active subreddits have a second Slack channel behind the scenes where they have communities where they discuss things.
You can even find John Mueller replied to me on one of my dark click studies because I posted in the SEO community on Slack, but not when I posted on Bluesky, not when I posted on Twitter, not when I posted on Facebook.
LUM KAMISHI
Yeah. Love that community.
JEREMY RIVERA
So there’s this fracturing, the splintering of the industry, because everybody was disgusted by what was going on at Twitter.
Some people fled to the Fediverse, they went to Mastodon, some went to Bluesky, some are on Threads.
But that diaspora has just created all of these different enclaves and hubs all over the place.
There isn’t just one place anymore.
But I feel like that’s the truth of our entire society.
We’ve shattered everything and spread it all out, and your literal world perception and experience happens through a shattering of different digital surfaces that then get algorithmically trained to what you kind of like, and then don’t surface the other stuff.
So we miss out on a lot of the challenge because we’re just shown what we’ll like.
LUM KAMISHI
Yeah, yeah.
So now the robots tell us what we should like.
And by the way, I love the SEO community.
Some really good discussions happened there.
But you reminded me of something.
I was looking at some old emails, and I saw an email from my friend.
He forwarded it to me, and there was this guy who owned like four websites in the gaming world.
And he just said, “I saw your comment on Moz. I’d like to connect with you and discuss how you could help me improve my sites.”
So this was a comment on a Moz blog, right?
And now, you’ve probably seen, I think that this was a good move by LinkedIn, where you can mark comments as AI slop.
I can’t even post there anymore, because I’ll have like four to five comments just copy-pasted into some LLM, and a regurgitated answer living in my comment section now.
So yeah, I totally feel you on this.
And maybe it’s a good thing that smaller communities now exist.
It has its downsides and upsides, but generally I think that’s good.
And there’s probably a few more Slack channels that I’m part of.
And the quality of discussions that happen there is no match for the public sphere, like the social media, the LinkedIns, the Facebooks.
I mean, Facebook is now complete garbage.
Even the groups, even the pages and everything is really, really rotten.
I even tried some Meta ads.
And it’s not just organic that’s broken, paid’s broken as well.
Yeah, we live in harsh times.
JEREMY RIVERA
Yeah.
Facebook groups: a nightmare for sellers, a dream for buyers
JEREMY RIVERA
You have to suss out and find certain things, and they only apply in certain use cases.
There’s one use case for Facebook pages, and it’s finding these really active community hubs, like in Cookeville here in Tennessee.
There’s Hip Cookeville, and man, if you want to get thirty-five different small businesses sending messages at you to fix your house, by all means post, “I want to build a deck,” and you’ll get seventy messages from every contractor, every indie builder, everybody with a license, without a license, and a bunch of DMs.
LUM KAMISHI
Deck builders.
JEREMY RIVERA
So if you do want a quick response, that does seem to be one thing that does still click.
That’s the only thing that I’ve seen: the home service niche does seem to have, for each geography, one active place where the older crowd are like, “Okay, where do I go to find information? I just want to post a question and have people pitch me their stuff.”
That’s Facebook groups.
So there is that one small sliver of light.
But again, I’ve only seen it work for home service or niche types of offerings.
And good luck being one of the sixty contractors messaging that poor person.
It’s like a one in sixty.
So I guess if you’re in early, maybe you’ve got a chance.
LUM KAMISHI
Well, if you’re a buyer, that’s a dream, right?
It’s a nightmare for sellers, it’s a dream for buyers.
What Lum hopes for: value over hype, and a Google that stops lying
JEREMY RIVERA
Yeah, that seems to be about the state of it.
What are you hopeful for in SEO in the next couple of years?
LUM KAMISHI
I’m really pessimistic about some things.
But I’m kind of hoping that this whole AI hype will somehow dry out and value will prevail.
Because I really think there’s too much AI slop there.
There’s an abundance of content, and we can’t see the line between what’s valuable and what’s not.
And I’m hoping that humans will train themselves and will find a way to distinguish between what’s slop and what’s not.
People in the industry, SEOs and everyone in the digital sphere, are already doing that.
But I’m hopeful for the larger population to get educated in that aspect as well.
And as for SEO, I just hope that Google will stop lying and will be more transparent.
I don’t want them to share their ranking factors with us, but I also don’t want them stealing clicks.
And this was a scandal really last week: the mock AI scraper scandal, that they were serving mock pages to AI scrapers.
That was really, really a petty move from them.
And so they’re making stuff really harder for everyone.
The quality of the results is not as good as it used to be.
Now we have AI Overviews, and I’m not really satisfied with what I’m seeing there either.
Probably they will get better with time.
I’m really skeptical that AI is going to take our jobs or overtake everything we’ve built so far.
But I really hope that Google shifts to a human-first search engine, right?
Because I really don’t think that AI Mode and the AI Overviews really serve the public, for that matter.
And of course it’s hurting SEOs with decreasing clicks, with decreasing traffic, especially publishers.
Publishers are the hardest hit.
Luckily I moved away from publishing long ago, so I didn’t get a taste of that in my own skin.
So yeah: more transparency, better results, better quality, better answers.
If you don’t have an answer, don’t just let your AI decide what the answer is.
And serve the public.
And don’t fight with SEOs.
Don’t pick a fight with SEOs, right? Let’s work together on this.
JEREMY RIVERA
I’m hopeful for that too.
I’m not holding my breath, but I can hope for that too.
LUM KAMISHI
Yeah, that’d be wrong.
This is wishful thinking, man. This is just wishful thinking.
I think that the situation will go from bad to worse.
It will get worse, but I hope, and I dream, that the other thing will prevail.
But I’m not hopeful, yeah.
AI slop and the uncanny valley
JEREMY RIVERA
What makes me hopeful is the kids in my homeschool tutorial.
I was flipping through videos to show what Oktoberfest is, and some Oktoberfest music came on, but it came with an AI video, and they’re like, “It’s AI, turn it off, it’s awful. Those are fake people.”
I’m like, “Yeah, it is awful, isn’t it, kids? This is AI slop.”
So we chanted “AI slop.”
LUM KAMISHI
There is this study I read way before ChatGPT came about, something called… what was the name of the phenomenon?
When the human brain perceives something as being…
JEREMY RIVERA
Uncanny valley.
LUM KAMISHI
Yeah, the uncanny valley, exactly.
When the human brain perceives something as being too human without it being human, it freaks out.
People in cartoons may seem cute, but the more they get closer to appearing human, the creepier they get.
So the majority of ads that I see now are AI people, right?
People that don’t exist, and it freaks me out, and I’m an adult.
So it’s really good that kids are getting freaked out about that as well.
So this is hopeful.
Human conversation is the raw material
JEREMY RIVERA
I think it drives the value up on human curation and human conversation.
My services have certainly evolved towards leveraging podcasts and leveraging podcasting content for maximum value, both in link building but also, certainly, this conversation.
This is my three, four, five thousand words’ worth of unique, relevant human conversation that is, as Matt Brooks of SEOteric puts it, human chow.
There isn’t something like an equivalent.
You couldn’t go to Semrush and keyword-research your way to what this article will be, the recap of our dialogue.
First off, ChatGPT is way too optimistic to say half of the things that we’ve said that are negative.
It’s just biased towards being positive.
You have to work really hard to get a prompt to give you a convincing, pessimistic view of stuff.
And I’ve also found it really difficult to build prompts.
I’ve done a couple of really good data studies recently, like analyzing backlink anchor text.
And the first draft of what Claude thinks is a good article about it is just so wishy-washy, doubly cautious about making any sort of statement.
I’m like, okay, well this is not going to work.
I’m going to have to take the reins and write most of this, because you’ve got to front your representations.
And I was wrong.
I thought my backlink study would show that exact-match anchor text still worked, or was the key way, and it turns out no, it’s token match optimization, similar to that bigger study that Ben Wills did.
If you token-match your anchor text, instead of asking for specific anchor text to happen every single time, it’s correlated almost two times better with having a positive impact.
But ChatGPT or Claude was afraid to make a summary statement like, “Hey, you’re doing backlink anchor text wrong.”
It won’t come out of the gate.
So when we’re talking to our clients and interviewing them, those should be recorded, transcribed, and then used as the framework that we feed these tools to craft great articles.
Almost 80% or more of the content I make is based on at least one or more transcripts from my interviews.
Any time I do SEO content, and every time that I do content work for clients, I try to get them on record and interview them, so that I have their voice, so that I have their opinions, and use that as a base that I transmute through Claude to create more content.
But I just find native AI stuff is just so wishy-washy.
LUM KAMISHI
Yeah, and it’s so obvious.
You can tell that it’s just wishy-washy.
I’ve never heard this phrase before.
They basically need a lot of food.
They need to be fed a lot and they need to be tweaked a lot.
I’ve used Claude and I’ve used AI tools to create content, and to be honest it’s been more of a challenge than to just write content myself.
I’m not really a good writer.
When I read my stuff, it lacks a certain flow, because I’m not native in English, and the tone lacks, and you can tell that I’m not native from writing and speaking and everything.
But I prefer my own writing to whatever Claude brings about from my prompts.
It takes a lot of effort to tweak it, and even after you generate articles and stuff, you always have to proof-check, you have to put your own hands in, and that takes time as well.
Previously we did a calculation.
We calculated the amount of time it takes a content writer to produce a 2,000-word article by hand, no AI, not for research, not for anything.
And the average was two articles per three days.
With Claude and everything, maybe I’ve been able to reduce that to two articles per day.
So we’re more efficient, but I’m not sure about quality at all.
That’s the biggest issue.
You can tell.
There are tells that humans can read and can pick up on.
And this whole thing with the watermark, I really don’t think that matters.
If a human can detect that, of course AI is going to detect it as well.
Lum’s time study for a 2,000-word article, by hand vs. with Claude.
Assertio.ai and how to reach Lum
JEREMY RIVERA
As a final question, give a shout-out for how people can find you.
If you’re building anything that you want people to go out and buy, if you’re available for consultation, let people know how they can contact you and what type of services you can give them.
LUM KAMISHI
Yeah, of course.
Well, you introduced me as Lum from Assertio.
That’s my SaaS project.
We talked a lot about AI slop.
So I built Assertio to tackle AI slop.
I’m still building it, it’s still in beta.
Assertio is a content creation pipeline that tries to automate content creation, but that’s why I know AI fails, because I know from my own experience.
And I’ve put a lot of effort into tweaking it into an answer-first content creation machine, and so far I’ve tested it on a few sites, and the results in terms of SEO seem to be good.
They seem to be better than what purely AI-generated content would look like, or human-only content would look like.
I would say its biggest strength is the research factor, because Assertio is a multi-agent pipeline, and one of the agents only does research, right?
It scours the web to find structure for the content at hand and to find information.
But as you said, you interview clients before writing content.
I write up their databases on my own, and I plug them into Assertio in JSON or CSV.
So we have their voice, we have their business information, everything they want to write about.
But in the end, I usually offer this as a white-label service, Assertio, because I still have to put my hands in it.
So it builds answer-first content.
If you remember, featured snippets are dead now, I think.
But most of my content ideas come from optimizing for featured snippets.
When we optimized for featured snippets, we did answer-first content based on the query: lists, ordered lists, unordered lists, paragraphs, tables.
And I’ve built all of this into Assertio.
So Assertio is like a projection of my whole decade-long experience in content and in everything.
People may reach me at lumkamishi.com.
I work primarily with marketplaces, e-commerce sites.
I’ve worked with businesses, local businesses, nationwide local businesses, SaaS companies, pretty much all of these guys.
But I mostly specialize in marketplaces and in large sites.
I’ve worked with sites that have billions of pages in their index, and my best work, I’d say, is content strategy, content creation, technical foundation building, and now of course AI search performance.
JEREMY RIVERA
Fantastic.
Well, thanks for stopping by and sharing a heartfelt view.
I’ll be providing you with some content from this interview if you want to get that published.
I’ll also be publishing versions of this transcript on unscriptedseo.com, seoarcade.com and my personal consulting site, jeremyriveraseo.com.
Thanks for stopping by.
LUM KAMISHI
Yeah, that’s cool.
Thanks for your time.
Thanks for having me.
It was a pleasure.
Key takeaways
- GEO is SEO plus digital PR. Lum treats GEO and AEO as an extension of good SEO: content, technical foundations and external citations. The extra weight sits on what the web says about you, meaning mentions from niche-relevant publications, linked or not.
- Niche-relevant links matter more now, not less. LLMs are more primitive than Google at ranking, so they are easier to move. Lum calls the comeback a “resurrection of links,” as long as the linking sites sit close to yours in vector space. Jeremy backed it with Ben Wills’ OppAlerts study: 20,000 positive matches and zero negative for topically aligned backlink profiles.
- Map every blog post to a service page. Lum scores “topical proximity,” the semantic distance between a post and the service page it supports, and builds mini clusters of two or three posts per service. Posts should help the money page rank without stealing its money queries. For one home-services client that meant killing 60% of the blog and promising 30+ new articles.
- Scope by results, not by retainer. The four-blogs-and-two-links agency package is gone in competitive niches. “I don’t care about budgets, I care about results” – it may still work in low-competition local or regional markets.
- AI drafts need feeding, and humans can tell. Claude cut Lum’s output from two 2,000-word articles per three days to two per day, but he is not sure about quality. Jeremy builds about 80% of his content from interview transcripts because raw AI output is “wishy-washy,” and his own anchor-text study found token-match anchors correlated almost two times better than exact match.
- Assertio is built as an answer-first pipeline. A multi-agent system with a dedicated research agent, fed client knowledge in JSON or CSV, structured the way featured-snippet content was: lists, tables and direct answers. Lum offers it as a white-label service while it is in beta.