How We’d Make $10K/Month From SEO Starting Today (From Scratch)

Edward Sturm

Abstract

Nearly two hours of practitioners answering a constrained question: keep the expertise, remove the assets, and reach ten thousand a month from a couple of hundred dollars. The answers are unsentimental — a matching domain in a niche small enough to understand, a network of cheap sites selling links, parasite pages catching whatever is spiking — and the second half turns into a working account of which platform signals are actually weighted and why.

• The asymmetry worth exploiting is geographic: a valuable term contested everywhere is uncontested somewhere.
• Bespoke automation should not be sold as a service, because it behaves like software.
• The risk with an autonomous agent is access rather than capability.
• Most platform algorithms are open or leaked, and the open ones give the weightings, which is the part that matters.
• Campaigns run at roughly nine parts human to one part machine, and the client approval step is never automated.
• The engine has shown it will let search quality fall for advertising revenue, and the earnings calls are the forecast.

Everything below follows the conversation in order, from the ten-thousand-a-month question through the testing culture and what old techniques are working again, into platform-by-platform signal weighting, communities, amplification and the answer engines, and out through the three-year forecast to who is actually winning. For anyone building a personal brand the useful part is how much of it is measurement: almost every claim here comes with the number behind it.

Chapter summaries

00:26 - The starting position is a small budget and a working knowledge of the field

Pick an exact-match domain in a small, low-competition niche you understand; sell what you are equipped to sell — yourself, or a local partnership where you rank someone's business and get into their leads.

"EMD some niche that I understand will rank a small site legend straight."

— 00:49

It is a deliberately constrained thought experiment: keep the expertise, remove the assets, and see what the operators actually reach for. What they reach for first is a domain that matches the search and a niche small enough to be understood.

01:31 - The niche should be hard to rank in, but only where the competitors are not specialists

Choose a high-value niche that is hard to rank for but only in the right geographies, so you are beating people who are not SEOs.

"I'll probably choose something in YM that is hard to rank for, but not in the right locations."

— 01:31

The asymmetry is geographic rather than technical. A valuable term can be contested everywhere and uncontested in particular places, and in those places the competition is businesses rather than specialists. Pick the fight you are equipped to win.

01:49 - One route is selling links off a network of cheap aged domains

Sell links: buy cheap aged domain names ($10-30), pump free AI content, boost domain rating with free links, keep publishing frequency up to maintain it; ~$30-50 setup per site, recoup on the first link sale, reinvest into a network of sites.

"I'm probably selling links to be fair, just straight up."

— 01:49

Cheap aged domains, generated content, authority raised with free links, and a publishing cadence that holds it. Thirty to fifty dollars a site, recouped on the first sale, and the proceeds go into the next site. It is a manufacturing process rather than a marketing one.

02:38 - Parasite pages catch whatever is spiking while the network builds

Alongside the link network, run free parasite pages to catch whatever niche is spiking — trendjacking, newsjacking, launchjacking — pick up clicks and high-ticket commissions, then reinvest into more sites.

"also at the same time be doing kind of free parasite pages as well"

— 02:38

While the network compounds, parasite pages catch whatever is spiking — trends, news, launches — for clicks and high-ticket commissions. The two run on different clocks, which is the point of running them together.

03:37 - The right content brings the outreach emails, and the sales land in month two or three

Put the right content out and you will immediately start receiving guest-post and outreach emails; sales usually come within month two or three.

"you'll immediately start getting guest post emails if you put the right content out there."

— 03:37

Publishing the right material produces inbound outreach almost immediately, and the revenue lands in month two or three. The lag between signal and money is the part most plans fail to budget for.

04:43 - Short buyer-intent pages outrank and outconvert the articles that answer questions

Most SEO writes articles to answer questions (how/what/when); a "compact keyword" approach instead puts up dozens of short pages that sell to searchers with buying intent — they rank and convert far better, and the average page is only ~415 words.

"This method of marketing is so effective, I had to make sure it wasn't against Google's rules"

— 04:43

Dozens of short pages selling to people with buying intent beat articles answering how and what. The average page is around four hundred words, which is a fraction of the work for a multiple of the return.

05:56 - Another route is selling the setup and the automations with a retainer behind them

Create lots of content on a video platform around AI/AI agents, then sell the setup and automations for ~$500-1,000, plus an ongoing retainer; five to ten sales and you are there.

"I was thinking maybe start a pyramid scheme. No, just just kidding."

— 05:56

Content about agents and automation, then the setup sold for several hundred to a thousand, with a retainer behind it. Five to ten sales reaches the target, which makes the arithmetic unusually legible.

07:16 - Bespoke automation should not be sold as a service, because it behaves like software

Do not sell AI automation as a bespoke service: it behaves like selling software — hard to scale, hard to bound scope, and may be irrelevant by delivery; sell a productised, tangible thing (links) or sell the community/course instead.

"for actually selling AI automation, I don't"

— 07:16

It behaves like selling software: scope is hard to bound, it does not scale, and the thing may be irrelevant by delivery. Selling a productised deliverable, or selling the community, avoids all three. Sell the thing, not the bespoke build.

08:57 - Testing means throwing everything at the engine, including the deliberately illogical

Test by throwing everything at the engine and seeing what sticks, including deliberately illogical tests to see whether the engine even spends the compute to catch them; re-run techniques from 2008-2011 — several work again now.

"maybe not strategy-wise because my strategies are pretty cleanly executed, you know, testing wise."

— 08:57

The method is to throw everything at the engine and watch, including deliberately illogical tests designed to find out whether the system even spends the compute to catch them. Techniques from fifteen years ago are being re-run, and several work again.

10:25 - A winning campaign is cohesive rather than a stack of separate tactics

Old techniques working again: redirect chains are the number-one thing working now; with enough authority you can stuff pages; JavaScript cloaking, once cleaned up, is ranking well again.

"Redirect chains, right, is probably the the number one thing that seems to be working at the moment"

— 10:25

On-page, publishing frequency, the link campaign, off-page and social all stacked at a natural pace. The cohesion is the strategy — individually each component is ordinary, and separately none of them moves anything.

11:11 - Old techniques are working again, and redirect chains lead the list

An autonomous AI agent's article beat a manual expert's better article for a competitive term and kept ranking, despite being obviously bad — high keyword density, many relevant entities, and a prompt that had worked before.

"you tried a lot of crazy stuff too."

— 11:11

Redirect chains lead the list of revived techniques. With enough authority, stuffing works again, and cloaked scripts, once cleaned up, are ranking. The revival is evidence about where the engine is currently spending its attention.

12:36 - A machine-written article beat a better human one and kept the position

A site with enormous domain authority ranks for everything, most of it unrelated to its topic — it becomes a "generic authority" whose pages rank across the board purely on backlinks and link profile.

"the domain authority of the website is like it's just so heavy."

— 12:36

An obviously bad machine-written article beat a better human one for a competitive term and held the position. High keyword density, a lot of relevant entities, and a prompt that had worked before — which is an uncomfortable result to report and worth reporting.

14:32 - A site with enormous authority ranks for everything, most of it unrelated

Give an autonomous agent a low-competition exact-match keyword and it writes a ~2,000-word SEO-optimised landing page, hosts it on a subdomain; you only buy the domain and add it to search console — near one-click publishing and ranking.

"The other experiment I've been doing recently"

— 14:32

Authority at a sufficient scale detaches from topic entirely. The site ranks for things it has nothing to do with, purely on its link profile, which is the clearest available demonstration that relevance is not the only gate.

15:51 - An agent given a low-competition term will write and host the page itself

For multi-domain brands, instead of redirecting every extension to one main site, build individual brand sites — more SERP real estate, more entity relationships, more visibility, more control over AI overviews.

"That actually makes a lot of sense for if you're trying to do like multi-dommain brand setups as well."

— 15:51

Given a low-competition exact-match term, an agent writes a two-thousand-word optimised page and hosts it on a subdomain. The human buys the domain and adds it to the console. Publishing has become close to a single action.

16:42 - Multi-domain brands do better with separate brand sites than with redirects

Feed an agent a system to build and develop its own topical map with the right retrieval pipeline and it will extend a site coherently; even manually approving each page builds big sites fast while controlling quality.

"I haven't been doing it with open club but I've been doing like through"

— 16:42

More results-page real estate, more entity relationships, more visibility and more control over what the answer engines assemble. Redirecting every extension into one site collapses all of that into a single position.

18:53 - An agent fed a retrieval pipeline will extend a site coherently

The real risk with autonomous agents is access, not capability — one sent crypto to the wrong wallet and lost ~$40k, another deleted an entire inbox despite being told not to; the human-in-the-loop is the executive who simply does not open the doors for Pandora.

"way worse situations. I think someone I saw one situation where someone uh got their open claw to send crypto"

— 18:53

Given a retrieval pipeline and a system for building its own topical map, an agent will extend a site coherently rather than randomly. Even approving each page by hand, the throughput builds large sites quickly while quality stays controlled.

21:19 - The risk with an autonomous agent is access rather than capability

Many platforms' ranking algorithms are open-sourced or leaked (a marketplace, a microblog, a short-video app, a professional network); the search engine's algorithm is far more advanced and dynamic, while the others carry linear signals that are much easier to manipulate.

"I've been trying in general to kind of reverse engineer a lot more algorithms."

— 21:19

One agent sent cryptocurrency to the wrong wallet and lost a large sum; another deleted an entire mailbox having been told not to. The capability was never the problem. The human in the loop is the one who does not open the door.

23:07 - Most platform algorithms are open or leaked, and the search engine's is not

On a microblog, a video attachment gets the maximum algorithmic boost, and there is a target interaction length — you want a ~3-5 second interaction, so keep the clip ~10-20 seconds; sentiment is also scored so the AI can process it.

"the very first post that I did was to promote that article, right?"

— 23:07

A marketplace, a microblog, a short-video app and a professional network all have open or leaked ranking systems, and they carry linear signals that are straightforward to manipulate. The search engine's is far more advanced and far more dynamic, which is why it resists the same treatment.

24:28 - A video attachment gets the maximum boost, and the target interaction is short

On that microblog, comments carry a roughly 75x multiplier and bookmarks outweigh likes, while a like is worth only one or two times.

"when you posted how important comments are now I make like a habit to comment on all of"

— 24:28

A video attachment earns the maximum boost, and there is a target interaction length behind it: three to five seconds of engagement, which means a clip of ten to twenty. Sentiment is scored as well, so the machine can process the reaction.

25:13 - Comments carry a large multiplier and bookmarks outweigh likes

Reach on a professional network collapsed after it was evangelised: 200k impressions a week became 1.4M by posting ten times a day, then fell to ~140k once everyone started posting daily.

"I think I ruined uh LinkedIn."

— 25:13

Comments carry something like a seventy-five-fold multiplier and bookmarks outweigh likes, while a like is worth one or two. The ordering tells you exactly which behaviours to design for, and most posting designs for the cheapest one.

26:10 - The open-source algorithms give the weightings, which is the part that matters

The professional network is the worst for AI spam comments; the way to crack it is one post outperforming that specific week — usually newsbait or a genuinely novel actionable post — and its impressions spread over a week, not the first hours, so consistency beats chasing virality.

"LinkedIn's probably the worst for AI comments as well in terms of replies."

— 26:10

Attributes without weightings are a puzzle; attributes with weightings are a specification. The leaks from the search engine gave the first, and the open-source platforms give the second, which is why the smaller platforms are more tractable.

28:18 - Reach on the professional network collapsed once everyone started posting daily

Where sales come from differs by platform: mostly video, microblog and paid social; the professional network drives serious, corporate and regulated niches (universities, local regulated practices) because its audience is CMOs and corporate networkers rather than operators like the speakers.

"Do you see uh many sales coming from LinkedIn? >> Oh yeah."

— 28:18

Two hundred thousand impressions a week became 1.4 million by posting ten times a day, then fell back to a hundred and forty thousand once everyone was posting daily. The tactic was destroyed by being shared, which is a lesson about tactics generally.

30:37 - The professional network is cracked by one post that outperforms that specific week

Signals tied to human effort are weighted above cheap ones: comments beat likes, and the microblog gives a 5-6x boost for replying to your own repliers inside a time window — forcing engagement and a Q&A dynamic.

"something that you said earlier about uh Twitter"

— 30:37

It is the worst platform for generated spam comments, and the way through is a single post that outperforms that particular week — newsbait, or something genuinely novel and actionable. Its impressions spread over a week rather than the first hours, so consistency beats chasing the spike.

32:01 - Where sales come from differs by platform, and the audiences are not the same

A community forum platform is best at catching AI spam because it analyses account behaviour over time, not a single post — a comment can stick and be auto-removed weeks later; an account behaves like a quality site (publishing frequency, karma, engagement) and dormant accounts get banned.

"effort versus human signals. I think the the best site we can look at so far"

— 32:01

Video, the microblog and paid social produce most of the sales. The professional network produces the serious, corporate and regulated work, because its audience is executives and corporate networkers rather than operators.

35:16 - Signals tied to human effort outweigh the cheap ones

Owning a branded community beats owning only a profile: you approve your own comments so they are never deleted, the sidebar links are do-follow (profile links are no-follow), and you can rank the community itself; take over inactive communities because new ones face a sandbox.

"created a subreddit at all and tried marketing on the platform for that? >> Yep."

— 35:16

Comments beat likes, and replying to your own repliers inside a time window earns a five- or six-fold boost. The pattern is consistent: the signal that costs effort is the one that counts, and the mechanism rewards a conversation rather than a broadcast.

36:44 - The forum platform catches spam by analysing behaviour over time

A new community faces a sandbox period, then explosive month-over-month growth driven by high-karma positive accounts crossing a threshold.

"we started a new one in January last year. You're right. There's like this sandbox period"

— 36:44

It analyses account behaviour over time rather than a single post, so a comment can stick and be removed weeks later. An account is judged the way a site is — publishing frequency, standing, engagement — and dormant accounts get banned.

37:31 - Moderators now read a summary of how an account behaves across the platform

Grow a community by posting all your content there, turning content into articles, and having the whole team post — more content pushed means faster growth.

"We just basically all our content we'll post on there and then we'll turn our content into articles"

— 37:31

Moderators now receive a generated summary of how an account engages across the whole platform, which moves enforcement from keyword recognition to behavioural pattern. It is a meaningful upgrade in what the platform can see.

40:01 - Owning a community beats owning a profile on somebody else's

After the forum-search partnership, the forum's algorithm barely changed but its anti-spam capability jumped; the transfer ran the other way — the search engine now takes direct ranking signals from forum up-votes, so out-upvoting an existing thread can outrank it.

"you see any changes on the maturity of the algorithm"

— 40:01

You approve your own comments so they are never removed, the sidebar links are followed where profile links are not, and the community itself can be ranked. Taking over an inactive one avoids the sandbox a new one faces.

42:06 - A new community faces a sandbox before the growth arrives

Build a custom AI project trained on your best-performing hooks and what flopped, then feed every new idea and transcript through it so output is optimised for what works; a typical day can then reach 500k+ people.

"I'll give you an example."

— 42:06

A new community waits, and then grows explosively month over month once enough high-standing positive accounts cross a threshold. Knowing the shape of the curve is what makes the waiting survivable.

42:51 - A community grows by everything the team publishes going into it

Do the same for video scripts: a trained project turns a trending topic into a one-click, ready-to-record script (third-grade language, strong hook, open loops, a mid-video CTA); a new channel hit 5k subs in a month and out-reached a 360k-sub main channel.

"we did the same with the scripts"

— 42:51

Everything the team publishes goes into it, content gets turned into articles, and everyone posts. The growth rate tracks the volume pushed, with no cleverness in between.

46:32 - The forum's anti-spam improved and the search engine now reads its up-votes

Open loops fight the drop-off: there is a measurable drop around the 4:30 mark, so plant an open loop there ("you're never going to believe what's next"); model the best creators' scripts and apply their techniques to your industry.

"Also, the open loops, we've been working on that a lot with the podcast."

— 46:32

After the partnership, the forum's ranking system barely changed while its spam detection jumped — and the transfer ran the other way too. The search engine now takes ranking signals from up-votes, so out-voting a thread can outrank it.

47:32 - A project trained on the operator's own hooks turns every idea into optimised output

The identical AI-content process that reaches 500k a day on an established account does not work at all on a brand-new account — momentum on the main account compounds.

"I don't know about whe you've seen this as well,"

— 47:32

A project trained on what performed and what flopped turns every new idea and transcript into output shaped by the evidence rather than by instinct. A working day at that level reaches upward of half a million people.

48:09 - The same treatment turns a trending topic into a ready-to-record script

Microblog amplification hack: if a post underperforms, have AI quote-retweet it (a 3x amplifier) with rewritten variations, across multiple accounts, until one hits ~10x; delete failures and respect the ~6-hour timing window.

"there's a hack that I wasn't going to cover, but I might as well"

— 48:09

The same treatment applied to scripts turns a trending topic into a ready-to-record draft: simple language, a strong hook, open loops, a call to action in the middle. A new channel reached five thousand subscribers in a month and out-reached a much larger one.

52:20 - An open loop fights the measurable drop-off partway through

Repurpose positive reviews onto social platforms in the language people use to search ("is this product a scam?"); people trust the same reviews more when found off-site, and social pages increasingly rank and get found.

"taking positive reviews and putting them with different language that people use to search for reviews of your product"

— 52:20

There is a measurable drop-off around the four-and-a-half-minute mark, so an open loop gets planted there deliberately. Modelling the best creators' scripts and importing their techniques into another industry is the general version of the move.

53:19 - The process that reaches half a million a day does nothing on a new account

LLMs are doing bigger query fan-outs and a growing number of grounding queries — after finding brands, the model runs more searches to do due diligence on them, so flood a brand's search results with positive content and the model finds it.

"we are seeing LLMs do much bigger query fanouts."

— 53:19

The identical process that reaches half a million a day on an established account does nothing at all on a new one. Momentum is doing more of the work than the method, which is worth knowing before copying the method.

55:21 - An underperforming post can be amplified by quoting it repeatedly

A "trials"/test-reel option pushes new content harder and was abused by re-uploading top reels; better to post one genuinely new piece a day; software and AI splice or reformat a video so the platform reads it as new and native per platform.

"Did you see Instagram's new real option?"

— 55:21

An underperforming post gets quoted repeatedly, in rewritten variations, across several accounts, until one lands — with failures deleted and a roughly six-hour window respected. The quotes point at the same original, because their impressions count toward it.

58:47 - Reviews get repurposed onto social in the language people search with

The first LLM query fan-out often uses a site: operator against known seed-list sites — if you have authority you have already been vetted and simply appear; the key is getting into the seed list, and a review-lookup is usually the second step where low reviews get cut from the answer.

"One thing I wanted to ask on the on the query find out"

— 58:47

The reviews get rewritten into the language people actually search with, including the sceptical phrasings. People trust the same review more when they find it somewhere else, and the off-site version does work the on-site one cannot.

59:58 - The models are running larger fan-outs and more grounding queries

The content types performing best inside the assistant are listicles (~19%), then SEO landing pages (~15%), then SEO product pages (~14%); sales pages built for SEO state why you fit the query, so the due-diligence fan-out finds your own argument — every company should carry heavy documentation.

"that bar in other niches I see as as a refinement"

— 59:57

Fan-outs are getting larger and grounding queries more numerous: having found a brand, the model runs further searches to check it. Flooding the brand's own results with positive material is how you decide what that check finds.

1:02:29 - A test option that pushes new content harder was abused by re-uploading

"AI will fix this" does not hold while the underlying blue-links algorithm still sources what the AI summarises — the output stays manipulable; it is only solved when the whole experience defaults to AI, and even then sourcing quality remains; as context gets cheaper the model pulls from more third-party publishers, so consensus favours link-builders, and you need ~100 pages to tune the output.

"I think literally Google's mentality is AI will fix this, right? But until it becomes the default,"

— 1:02:29

A test option that pushed new content harder was immediately abused by re-uploading top performers. Posting one genuinely new thing a day is the durable version, with tooling used to reformat it so each platform reads it as native.

1:04:45 - The first fan-out often runs against a seed list a site is either on or not

To predict the search engine three years out, read the company's earnings calls — they do not lie about revenue goals; the last two years have been reactionary (code-red versus AI rivals), and the trajectory follows wherever they see ad dollars.

"what do you think SEO looks like in three years from now?"

— 1:04:45

The first fan-out often runs against a known seed list, and a site with authority has effectively been pre-vetted and simply appears. Getting onto the list is the real objective, and a review check is usually the second step, where low ratings are cut from the answer.

1:07:10 - Listicles perform best inside the assistant, then landing pages, then product pages

The engine has shown, in leaked internal communications around a major update, a willingness to let search quality drop for ad revenue; trying to serve everyone with AI while rivals narrowed and pulled ahead risks degrading organic results and disconnecting source quality.

"One thing that I've been thinking about right"

— 1:07:10

Listicles lead at around a fifth, then optimised landing pages, then product pages. A sales page written for search states why it fits the query, which means the due-diligence pass finds your own argument — an argument for carrying heavy documentation.

1:10:01 - The providers are burning cash, so the loopholes stay open a while

The "helpful content" update was really market consolidation, not a quality move — the winners were big forums, video and Q&A platforms, the losers independent publishers and experts; the same content ranks on the big platforms but not on the expert's own blog.

"why you're trying to go down this path either."

— 1:10:01

The providers are burning cash and fighting to upgrade, so search quality is not where the attention is. The practical consequence is that loopholes and revived techniques stay open for a while longer than anyone expects.

1:12:46 - "The model will fix this" does not hold while it still sources from the old index

Keep high-level campaigns manual and run roughly 90% human / 10% AI: automate guest-post content and topic research, but have the team do the work and then ask the AI what they missed; AI-augmented workers fly, workers who let AI do the whole job are obvious and get fired fast.

"what aspects of your SEO are you automating?"

— 1:12:46

While the underlying index still supplies what gets summarised, the output remains manipulable. It is only closed when the whole experience defaults to the model — and even then sourcing quality decides it. As context gets cheaper the model pulls from more publishers, which favours whoever can place material across many.

1:14:39 - To predict the engine three years out, read the earnings calls

Automate freely for your own properties, but for clients the quality-approval step is the big manual cost — anchor text, the insertion sentence, and any on-site content must be human-checked, because you would rather catch it than the client does.

"saw you shake your head when uh"

— 1:14:39

Earnings calls do not lie about revenue goals, which makes them a better forecasting instrument than any commentary. The last two years have been reactionary, and the trajectory follows wherever the advertising money is seen to be.

1:16:22 - The engine has shown it will let quality drop for advertising revenue

Build retrieval "brand dossiers" that aggregate every data point for a case — search-console and link exports, goals, stakeholder context — so SEO decisions align with everyone's interests; recommending an H1 change the brand will reject makes you look a fool and loses the SEO benefit anyway.

"I got a a a new obsession around rack pipelines"

— 1:16:22

Internal communications around a major update showed a willingness to let search quality fall for advertising revenue. Trying to serve everyone while rivals narrowed and pulled ahead risks degrading the results and disconnecting them from source quality.

1:19:05 - Whoever wins the model race, the compute layer wins regardless

Read patents at scale: define once, precisely, how you want a patent analysed, then have the AI read 150 and explain each simply; sort them into content, technical and link-building, and bake the validated mechanisms (heading vectorisation, anchor placement, image placement, passage ranking) into your content.

"I've been, you know, collecting Google patents for the longest freaking time."

— 1:19:05

The infrastructure layer sells hardware and power to every outcome, so it wins whichever model wins. Several of the model providers are burning cash and are forecast to run low, which makes the compute layer the safest position in the whole race.

1:30:40 - The helpful-content update was consolidation rather than a quality move

The operators winning have sober thinking and know how to do both: on their own money they test everything, on entrusted projects they do only what is rock-solid enough to last ten years; do not get engulfed by the GEO noise — it is SEO done well, just with different variables weighted differently.

"I'd say those I'm extremely biased. Those who have very sober thinking"

— 1:30:40

The winners were large forums, video platforms and question-and-answer sites; the losers were independent publishers and experts. The same content ranks on the big platforms and not on the specialist's own blog, which is a consolidation outcome rather than a quality one.

1:31:46 - High-level campaigns stay manual at roughly nine parts human to one part machine

The frame is not white-hat versus black-hat — that implies an intention to cheat — it is leverage: AI plus the right systems, and testing enough to know what objectively works, then applying it.

"I mean it depends what you define as white hat right"

— 1:31:46

Roughly nine parts human to one part machine. Topic research and guest-post drafting get automated; the team does the work and then asks the model what was missed. Augmented workers fly, and workers who hand over the whole job are obvious and do not last.

1:33:03 - Automation is free on owned properties and forbidden on the client approval step

The biggest corporate white-hat brands win their giant keywords because real-world signals amplify their digital ones and compound; everywhere else — the non-giant SERPs — the black hats win.

"it depends on the setup. Number one, um the very first thing is that the"

— 1:33:03

On your own properties, automate anything. For clients, the approval step is the expensive manual cost and stays manual: anchor text, the insertion sentence, any on-site content. You would rather catch the error than have the client catch it.

1:33:50 - A brand dossier aligns the recommendation with what the brand will actually accept

A white-hat expose of a black-hat agency was itself removed via a DMCA/negative-SEO takedown and deindexed for 48 hours, reindexed only after public outcry — black hats now use the engine's own enforcement actions to remove competitors.

"a fantastic example of this,"

— 1:33:50

Every data point for a case in one place — console and link exports, goals, stakeholder context — so the recommendation aligns with what the organisation will actually accept. Recommending a change the brand will reject wastes the recommendation and the credibility.

1:35:35 - Patents can be read at scale once the analysis is defined precisely

Outside giant-brand keywords, the best black hat outranks the best white hat in weeks versus a year; by the time the white hat ranks, the black hat has compounded the traffic, profit and capital — so take the algorithm's results with a grain of salt.

"if you're wanting to go into any SER, if I was to put the best black hat in the"

— 1:35:35

Define once, precisely, how a patent should be analysed, then have the model read a hundred and fifty and explain each simply. Sort them into content, technical and links, and bake the validated mechanisms into the way pages get built.

1:37:02 - The people worth learning from are the reason to be wherever they are

Even if AI mode becomes the default, the multi-publisher off-page operator expands rather than loses — one website does nothing, but building across 100 publishers to manipulate the end output beats owning a single default experience, and technical on-page competitors get removed.

"even if it becomes the default, my biggest investment side, we actually expand"

— 1:37:02

The answer to why so many operators are in one place was simply that the others were there. It is circular and it is correct: proximity to the people worth learning from is the input, and everything else follows from it.

1:37:41 - The operators winning are the ones testing on their own money and shipping only what lasts

AI mode as default is inevitable — the engine's earnings calls targeted Q4 — but call timelines are grandiose and unreliable (a five-year-old AI-calling demo still has not shipped); it is coming because it keeps users on-site and fewer clicks make each remaining ad worth more.

"I do see it happening, right?"

— 1:37:41

On their own money they test everything; on entrusted projects they do only what is solid enough to last a decade. The framing that matters is leverage rather than white against black, because the latter implies an intention to cheat that is not what is actually being described.

1:39:28 - The frame is leverage rather than white hat against black hat

Position yourself one notch above where you are, ahead of each update: prior updates that wiped niche blogs and affiliate sites were foreseen, so move from consultant/affiliate toward corporate AI/GEO expert taking six-figure contracts before the change arrives.

"Well, I try to I I've done the exact same thing with HCU."

— 1:39:28

Real-world signals amplify the digital ones for the largest brands, and they compound, so the giant terms stay theirs. Everywhere else the aggressive operators win — often in weeks against a year, which means the visible results should be read with that in mind.

1:40:47 - The engine takes the cake unless it is a new cake it wants to hand out

The engine takes your cake unless it is a new cake it loves handing out — sites demoted for being publicly about SEO were later featured once the operator talked about the engine's own AI model.

"it's so funny. I was going to say it's so funny you say about"

— 1:40:47

Platforms reward what promotes their current priorities and demote what threatens them, so the topic that gets amplified is the one the platform wants amplified. Sites demoted for covering the discipline were later featured for covering the engine's own model — told as a personal experience rather than as a measured pattern.

Personal Branding Lessons

A conversation between operators comparing notes on what they have measured. The moves below are the ones that transfer out of their context.

Pick a niche that is hard to rank in only where the competitors are not specialists

The asymmetry is geographic rather than technical. A high-value term can be genuinely contested in some places and barely defended in others, and in the second kind you are competing against businesses rather than against people who do this for a living. Choosing that ground is most of the advantage. 01:31

Never sell bespoke automation as a service

It behaves like selling software: the scope resists being bounded, it does not scale the way a service should, and the thing may be irrelevant by the time it is delivered. A productised deliverable avoids all three, and so does selling the community or the course instead of the build. 07:16

Keep an autonomous agent away from the doors rather than away from the work

One agent sent funds to the wrong destination and lost a large sum; another deleted an entire mailbox having been explicitly told not to. Neither failure was a capability failure. The human in the loop is not a reviewer of output — it is the person who simply does not grant the access. 18:53

Own the community rather than the profile

On a community platform, owning the space means approving your own comments so they are never removed, sidebar links that are followed where profile links are not, and the ability to rank the community itself. Taking over an inactive one avoids the waiting period a new one faces. 35:16

Train a project on your own hooks and run everything through it

A project trained on what performed and what flopped turns every new idea into output shaped by evidence rather than instinct. The same treatment applied to scripts produces a ready-to-record draft from a trending topic — simple language, a strong hook, open loops, a call to action in the middle. 42:06

Run campaigns at nine parts human to one part machine

Topic research and drafting get automated; the high-level campaign does not. The team does the work and then asks the model what was missed, which is the inversion that matters. Augmented workers move fast, and the ones who hand over the whole job are obvious and do not last. 1:12:46

Build a brand dossier before recommending anything

Console exports, link data, goals, stakeholder context — aggregated once, so that the recommendation aligns with what the organisation will actually accept. Recommending a change the brand will reject costs the recommendation and the credibility, and the benefit never arrives either way. 1:16:22

Position yourself one notch above where you are, ahead of the update

The updates that wiped niche blogs and affiliate sites were foreseeable, and the people who moved before them kept their income. The move is to be one step further up the value chain than your current work requires — consultant toward specialist, specialist toward the contract nobody else can service. 1:39:28

Questions

Each answer ends at the moment in the recording where it is given.

If I were starting from scratch today with almost no money, what would actually get me to $10k a month?

The routes offered all attach a ranking to demand that already exists. Rank a small, well-understood niche and plug it into a local business that needs leads, or target a lucrative category only where its competitors are not optimisers. A second path builds link inventory on cheap aged domains, recouping the small setup on the first sale and reinvesting into a network. The common thread is monetising where ranking meets an existing buyer, not building appetite from nothing — often in a narrow niche whose economics favour the small. 00:49

Why does obviously bad AI content sometimes outrank genuinely better content?

Because authority is doing more of the work than quality right now. A thin, keyword-dense agent article beat a genuinely better human one and kept ranking, sitting on a heavy domain with the right entities and a prompt that had worked before. Scaled up, a site with enormous accumulated authority becomes a generic authority that ranks across unrelated topics purely on its backlink profile. Past a threshold, any page matching some level of intent ranks regardless of topical fit — offered as a puzzle, not a settled explanation. 11:11

How are people using autonomous AI agents to build and rank whole sites?

Given a low-competition exact-match keyword, an agent can write a full landing page, host it on a subdomain, and leave only the domain purchase to a human — near one-click publishing. Fed a retrieval pipeline and a scheme to build its own topical map, it extends a site coherently, with page-by-page human approval acting as the quality gate. The described boundary is access, not capability: agents are kept self-contained with extremely limited permissions, because the real damage comes from broad access, not bad drafting. 14:32

Which signals do the social and forum algorithms actually reward most?

The signals hardest to fake. On one microblog a video attachment earned the maximum boost, comments carried a multiplier many times a like, and replying to your own repliers inside a window multiplied reach again. A community forum judged accounts on behaviour over time rather than a single post, treating an account like a whole site. The recurring lesson is that effort-heavy signals beat cheap ones, which is why knowing a platform's weightings matters more than knowing its attributes — and why where you choose to build shapes what you can win. 21:19

How do I get an AI assistant to recommend my brand?

Assistants increasingly run a second wave of grounding searches to vet the brands their first search surfaces, and often lean on a vetted seed list where authority means you simply appear. So the play is to flood your brand's own search results with positive content in the language people actually search, seed proof off-site where the model will find it, and build enough documentation that a due-diligence search finds your own argument for why you fit. Low-review, thin presences get cut from the answer for lack of confidence. 53:19

What will search look like in three years, and how would I even predict it?

The method offered is to read the company's earnings calls, because a public company does not lie about revenue goals, and stated targets for search-ad growth have to come from organic results. The recent trajectory is described as reactionary, chasing AI rivals, and the direction is said to follow wherever the ad dollars are — even as announced timelines slip by years. The forecast is that the experience keeps moving toward AI answers because they raise time on site and make each remaining ad worth more. 1:04:45

Are the black hats or the white hats winning right now?

The answer rejects the label and lands on temperament: those with sober thinking who test aggressively on their own money and do only what is rock-solid on entrusted work. The largest corporate brands win their own giant keywords because real-world signals amplify their digital ones, but everywhere else the faster operator reaches the top in weeks against a year and compounds the lead. Judged by what objectively works rather than a moral filter, speed-to-rank is banking the advantage — offered as a biased, personal reading, and worth taking with a grain of salt. Genuine leverage is the frame that survives. 1:30:40

How do I stay ahead of the next algorithm update instead of being wiped by it?

By repositioning one tier above where you are before the change arrives. Earlier updates that wiped niche blogs and affiliate sites were foreseeable to anyone watching the engine's intent, and the response is to move early — from the tier about to be devalued toward the one about to be rewarded — rather than defend a shrinking patch. The governing idea is that the platform reclaims whatever value you hold unless it is value it currently wants to promote, so aligning your topic with what it is amplifying is the protection, a move closer to shifting position without starting over. 1:39:28

Sources

How We’d Make $10K/Month From SEO Starting Today (From Scratch)

The sections above follow the episode's own order. Each timestamp is the point in the recording where the idea is discussed, so the post reads as a map of the conversation rather than a rearrangement of it.

Quotations come from the episode transcript with spoken filler ("okay", "right", "you know", "um", "actually", "kind of", "sort of") and false starts removed. No words are added: square brackets mark anything inserted for sense, and an ellipsis marks any cut. British spellings are restored, and speech-to-text errors were repaired against context only where the intended word is unambiguous. Where a line appears both in a cold-open teaser and in its place in the conversation, the timestamp cites the second — the point at which it is actually said.

Quotes are stamped with the moment they are said rather than with a speaker's name.