THIS IS THE ARTICLE IS WAS NOT GOING TO PUBLISH. It's the most in-depth course you will get on doing SEO with AI (and i'm sharing it all for free).

TLDR:

1. How to pick the keywords that pay.
2. Build pages better than the top 3.
3. Earn links, free first and paid last.
4. Get named by others so ChatGPT cites you.
5. Repeat every week, because the order matters more than any single tactic.

So I took everything I've published, cut what aged, fixed what I got wrong, and put it in the order we actually run it. Last thing, if you want me to continue doing these, I need your support. So a like, a quote or a bookmark would help me continue 🙏 Back to the course 🔥

Before we start: what AI can and cannot do for your SEO

AI does not know your search volumes. It does not know your rankings.

Ask Claude which keyword to target and it guesses from its training. Ask it where your site ranks and it cannot know. Everything downstream of a guess is a guess.

What Claude is great at:

• Reading Search Console exports and finding the patterns you would miss
• Writing 30 pages that follow one structure without drifting
• Adding internal links to every old page you own, which no human does twice
• Writing structured data, titles and meta descriptions at scale
• Running the same check every Monday without forgetting

What it cannot know on its own:

• Search volume and keyword difficulty
• Your rankings, clicks and impressions
• What Google shows on the results page today
• Which competitor pages are taking your traffic

So the job splits in three.

• Claude is the hands. Opus 5.5 in Claude Code if you have a repo. Sonnet 5.5 in a Claude Project if you don't. It plans, writes, edits, runs the routines.
• ChatSEO is the brain. Through its MCP it brings your Search Console, keyword volumes, the live SERP, competitor pages scraped to the source code, and a process built from how a consultant actually works.
• You approve. Anything that touches a live page, or leaves under your name.

Setup takes five minutes. Create a ChatSEO account, add your site, connect Search Console. Skip that last step and half of what follows runs blind. Then in your terminal:

claude mcp add --transport http chatseo https://api.chatseo.app/mcp

Open Claude Code, type /mcp to log in once. In the Claude app, add https://api.chatseo.app/mcp as a connector instead (the MCP is on the Pro plan).

Every prompt below that says "Using ChatSEO" assumes this connection. The ones that say "skill" are files: save them as .claude/skills/<name>/SKILL.md in your repo and Claude Code picks them up.

One more thing before Part 1.

GEO is not a separate discipline you start after SEO. When ChatGPT doesn't know the answer, it searches Google. If you have no SEO, you have nothing for the model to find.

-> Good GEO needs great SEO first. That is why GEO is Part 4 and not Part 1.

Part 1: How to pick the keywords that pay

Most SEO traffic is vanity.

"What is SEO". "CBD benefits". Your Search Console curve goes up, you feel great, and nobody buys.

Paul, my co-founder, puts it bluntly: is that traffic for your ego or for your business?

Module 1: Target the card in hand

The only keywords worth your first six months are the ones where the searcher already has the card in hand.

"SEO consultant Paris". "Coffee table". "Shopify SEO agency". "Alternative to Mangools".

You do not need a tool to read intent. Google already did the testing. Type the keyword and look at what ranks.

• Blog posts and a definition box at the top: informational. Skip for now.
• Landing pages, service pages, category pages with prices: commercial. That is your target.

Low volume is not the price of admission. "[Competitor] alternative" might do 90 searches a month. All 90 of those people are shopping for a replacement today.

Module 2: A keyword gets you a page. A pattern gets you a machine.

A pattern is a set of keywords with identical intent, where Google returns the same type of page every time. Only one word changes.

• "[Tool A] vs [Tool B]"
• "Best SEO tool for [ecommerce]"
• "SEO consultant [city]"
• "[Service] [city]"
• "[Product type] [material]"

One pattern is dozens of pages. You never pick a topic again. You pick the next row.

The free way to find them is Google Suggest. Type your seed, then add a * at the end, then at the start. Each one surfaces a different batch. Google Ads' Keyword Planner adds the volumes.

The fast way is to let Claude ask the analyst.

---
name: money-keywords
description: Find the commercial keyword patterns for this site. Use when asked which keywords to target, for an SEO strategy, or "where is the money".
---
1. Call ChatSEO list_sites and pick this site's ID.
2. Send ChatSEO (send_message, with the siteId), word for word:
   "Give me the best SEO strategy for my site. Group the opportunities by keyword pattern, and only keep the ones with commercial intent. Rank the patterns by how realistic they are for a site with my authority."
3. Poll get_conversation_messages until the reply is complete.
4. For each pattern, keep only keywords where the current top 3 are landing, service, category or product pages. Drop anything where blog posts dominate.
5. Write keywords.md: pattern, keyword, monthly volume, page type Google ranks, and whether we already have a URL for it.
6. Never invent a volume. If ChatSEO did not return it, write "n/a".

Step 6 is there for a reason. Claude will happily fill a blank column with a plausible number. The analyst exists so it never has to.

Module 3: If you sell software, there are 9 page types. Write them in this order.

The order is by intent and by how hard the SERP is.

1. Alternative pages. Alternative to X, free alternative to X, cheaper alternative to X. Highest intent on the list. No company writes a page helping you leave them, so the incumbent cannot compete. Write these first.
2. Comparison pages. X vs Y, X or Y, X vs Y vs Z. Two vendors and a pile of affiliate content is the whole field. Be honest about the rows where you lose. It is the only thing that makes the rest believable.
3. Integration pages. Your tool + Slack, + Stripe, + Shopify. You borrow the search volume of somebody else's brand and almost nobody fights you for it. Twelve integrations is twelve pages.
4. Constraint pages. Without subscription, without credit card, open source. Binary filters, not opinions. If you match the constraint, you win the page outright. If you don't, don't write it.
5. Pricing pages. People search price before product. Directories already sit on most of these, so do them once you have some authority.
6. Use case pages. Tools to generate leads, to automate prospecting. You rank for the job instead of the category. Slower to land.
7. Problem pages. The reader doesn't know your category exists yet. Weak competition, low intent, real payoff at month six.
8. Feature pages. Only for features you genuinely lead on. Otherwise you are publishing a directory listing about yourself.
9. Company size pages. For freelancers, for SMEs. I would mostly skip these. They are almost always the same page with three words swapped, and Google has a name for that pattern.

Local business? Your patterns are "[service] [city]" and "[category] [city]". Part 2 has the full local system.

Module 4: Pick the SERPs you can actually win

Most beginners burn their first months on keywords they can't rank for.

Two filters, and the first one kills most projects:

• The SERP is beatable at your current authority. We could not rank "Semrush alternative" when we started. Too competitive. Versus SERPs were thinner, and the reader was already comparing tools with a card in hand. That trade was deliberate.
• The intent is commercial. Module 1.

On a young site, go small and specific. A plumber in Austin should not start on "plumber Austin". The list underneath is where you win first: "emergency plumber Round Rock", "water heater repair Cedar Park". The suburbs pay the same rates as downtown.

The mistake that cost us the first months

Before ChatSEO, Paul and I ran a business plan generator.

The SEO opportunity was real. Thousands of programmatic queries. Low competition. Obvious buyer intent.

So we launched fast and published fast.

It never ranked. Not slowly. Never.

The pages were not the problem. A brand new domain, zero brand searches, zero backlinks, and a publishing rate that looked exactly like every site Google files under "generated at volume".

-> We were nobody, showing up loud. Google had no reason to spend crawl budget finding out whether we were the exception.

Everything in Parts 2 to 5 exists so you don't repeat that.

Part 2: Build pages better than the top 3

The most expensive SEO mistake is writing a page you already had.

The second most expensive is writing one that adds nothing to the nine results already there.

Module 5: One keyword, one page. Decide before you write.

Here is the check Paul does by hand. Search Console, Performance, filter on the exact query, then the Pages tab.

• A page already gets clicks for it: optimise that page.
• A page gets impressions but few clicks: it is already somewhere in the top 10. Optimise it.
• Nothing: create a new page.

Get this wrong and you build cannibalisation. Two of your pages fight for the same query and Google cannot choose. Paul's example: both sit 7th and 8th, when one page would have been 3rd.

Then the part most consultants skip: secondary keywords. Take whoever ranks 1st for your keyword and list every other query that exact URL ranks top 5 for. That list is what Google already considers the same page. Target it all with one URL.

It all lands in the title.

The title is the page's identity card, the strongest on-page element you have. One word decides whether a page ranks for one query or five. "SEO Consultant Paris (Freelance)" covers two patterns at once. If pixels remain, spend them on the click.

The meta description barely moves rankings. Its job is to sell the click.

---
name: page-brief
description: Build the SEO brief for one target keyword. Use when asked to rank for a keyword, optimise a page, or before creating any new page.
---
1. Send ChatSEO (send_message, siteId), word for word:
   "I want to rank first on <keyword>. Read the live top 10. Tell me which page type wins, which secondary keywords the top 3 also rank for, and whether I should optimise an existing page or create a new one. Then give me the full brief."
2. Poll until complete. Extract: create or optimise and which URL, primary and secondary keywords, the sections the top 3 all have, the title and meta description proposed.
3. If an existing URL already gets clicks or impressions for the keyword, STOP any plan to create a new page. We optimise.
4. Check the title contains the primary keyword and fits without truncating in a SERP preview.
5. Save the brief to briefs/<keyword-slug>.md. Do not edit the site yet.

When Paul ran this live on his Lyon consultant page, ChatSEO flagged a canonical tag he had never noticed. He has been doing SEO since 2021. That is what the analyst is for.

Module 6: Build what the top 3 are almost building, plus one thing they don't have

Open the top 3 and put yourself in the searcher's head.

Someone typing "SEO consultant Paris" wants to know why you, what it costs, how it works, and how long until results. Answer those, in the order the SERP answers them, and you are level.

Then do it better. Paul's teardown of a competitor ranking 1st: no photo of the person you would hire, no way to self-qualify. Add the face. Add a three-question quiz.

That last part is not decoration. It is your spam defence.

In March 2024 Google added scaled content abuse to its spam policies, and the wording matters. It applies no matter whether the content came from automation, humans, or a mix of both. What it targets is large amounts of unoriginal content that gives readers little to no value.

-> Volume is not a violation. Unoriginal content is.

Google's guidance on AI content says the same from the other side. They "focus on the quality of content, rather than how content is produced".

So for every page, one question: what is here that the other nine results do not have?

• A quiz that routes the visitor to the right answer
• A calculator or a simulator
• Your own data, or a table with numbers you actually pulled
• Real photos of the job, the team, the place
• Video

One of them on every page, or do not ship the page.

The rest of on-page is mostly noise:

• One H1 with your primary keyword. The H2 and H3 tree takes care of itself.
• Structured data. One instruction to Claude at publish time. Agencies hate it because half the fields only exist once the page is live, final URL, publish date, author. Claude publishes the page, so it has all of them.
• Internal links are a threshold, not a lever. Google discovers pages by following every link from your homepage. A page nothing links to barely exists. Paul's floor is five internal links to every new page, from pages on a close topic, inside body text.

-> A link in a sentence tells Google what the next page is about. A link in a menu tells it nothing.

Module 7: Give Claude a template and a process file, not a prompt

Left alone, Claude ships black text on a white background.

Technically a page.

Step 1: Rebuild one page you already like as the template

Open a page from your own site. View source. Paste it into Claude Code:

Rebuild this page as the template for every SEO page on this site:
same design, title pattern, meta description, H1, FAQ block,
interactive element and CTA placement. Save it and use it for every
new page.

One shot the first time, two or three fixes after. Works even if the page was built in Elementor. Under the hood it's all code, and Claude reads code.

Step 2: Write the SEO process as a file

A markdown file Claude reads before every page. Two lines per rule is enough. Ours:

# SEO process: read before every page
- Build on the template. Never invent a layout.
- Add at least 5 internal links to the new page, from the body of related pages. Anchors match the target keyword of the new page.
- Translate to French and link the two with hreflang.
- Add the structured data that matches what is actually on the page.
- Update the sitemap on every publish.
- Request indexing in Search Console.

The internal links rule is the one that pays. Claude can open every page you own, including the old ones, and add the links itself. A human never does that consistently. I certainly did not.

The sitemap trap

The second you let Claude publish, you have left your CMS.

Paul found out the hard way. He was testing a ChatSEO feature that reads a site's sitemap, and it kept ignoring a page he had published the day before. The page was live. The sitemap had not moved since he stopped relying on his plugin.

Nothing warns you. Your site just quietly stops telling Google what exists. That is why "update the sitemap" is in the file.

The day Claude deleted the one thing that mattered

We asked Claude Code to duplicate one of our comparison pages for four new competitors.

Everything transferred. Titles, meta descriptions, structured data, table of contents, the FR and EN pair.

The quiz was gone. Both pages. No error, no warning, nothing in the output.

The quiz was the one original element. The exact thing the scaled content policy is written about.

Then it happened to Paul. Claude rewrote his city pages and dropped the "best SEO consultants" ranking tables. He found out live, recording a podcast.

An agent that can edit your site can quietly remove the one thing the page existed for. So our ship skill has a gate.

---
name: ship-page
description: Write or update a page from its brief and publish it safely. Use after page-brief, when asked to build, rewrite or publish a page.
---
1. Read briefs/<keyword-slug>.md. Write or rewrite the page on the template, covering every section the top 3 share, plus one thing they do not have (quiz, calculator, photos, real numbers, video).
2. One H1 with the primary keyword. Title and meta from the brief.
3. Follow the SEO process file.
4. DIFF GATE. Before publishing, list every element present in the old version (or the reference page) and missing from the new one: tables, rankings, forms, quizzes, embeds, images, internal links. Show me that list. If it is not empty, say so in the first line.
5. Publish only after I say yes. Draft or branch first, never straight live.

Step 4 is the whole lesson. The day it catches nothing is the day you stop needing it, and that day does not come.

Module 8: Duplicate the pattern, slowly

Once one page in a pattern is right, the next forty are the same page.

Consultant + Lyon has the same intent as consultant + Lille. So you build page one by hand, take the full hour, and duplicate from there.

You are about to multiply this page by twenty. Every weakness multiplies with it.

The trap: a page where only the city name changes and nothing else does. That is a template with a find-and-replace. Google's spam policy calls it doorway abuse, and its example is literally "pages targeted at specific regions or cities that funnel users to one page". Every variant needs something only that variant can say. City facts, prices, examples, testimonials.

The cadence is the strategy

Our blog runs on one sentence, twice a week:

Take the last article we published, and do the next keyword in the pattern.

Claude Code checks which variants exist, picks the next one, gets the brief and the article from ChatSEO, builds it on the template, translates it, links it, updates the sitemap and asks Google to index it.

2 articles in English, 2 in French. Four pages a week. A person reads every one before it goes live.

A year ago the ChatSEO blog got close to zero visits from Google. Today it gets about 3,000 a month.

The curve was not a straight line. It climbed to about 1,500 in the spring, fell by half, then came back to 3,000. We kept publishing 2 a week through all of it.

-> A site that publishes 30 pages on Monday and nothing for a month looks like a content bomb. A site that publishes 2 every week for a year looks like an editorial team.

If I were Google and wanted to separate editorial teams from people bombing content, publishing volatility is the first thing I would look at. Pick a number. Hold it. One page a day is the ceiling on a new site.

Going multilingual

The hour is already paid. A second language costs almost nothing.

Check your domain first. On a .fr, a .de or a .co.uk you are not doing multilingual SEO, no matter how clean the implementation. Paul tested it on his .fr. The country domain is too strong a signal. Use a .com with subdirectories.

Then two checks: a language switcher that goes to the equivalent URL and not the homepage, and correct html lang plus hreflang. Get hreflang wrong and your two versions compete with each other instead of covering two markets.

Module 9: Before you publish more, fix what you have

The fastest traffic is already on your site.

On a 50-page blog I went from 1,500 to 15,000 monthly clicks in three months without publishing a single new article. Three moves, in this order.

Move 1: Cut 15 to 20% of the pages

Pages that sat there for months with no clicks, written because I thought I should be publishing. Every one diluted the signal about what the site was about.

The filter: zero clicks and negligible impressions over 90 days, excluding anything published in the last three months. New pages haven't had time to prove anything. Cutting them is impatience.

Using ChatSEO, show me every page with zero clicks and under 50
impressions in the last 90 days, excluding anything published in the
last 3 months. For each one, tell me whether I should delete it,
merge it into another page, or rewrite it.

Move 2: Repoint pages at the keywords Google already gives them

This moved the most traffic, and almost nobody does it.

For each page, look at which queries it actually gets impressions for. Not the keyword you wrote it for. The keyword Google decided it was about. On most pages those were different.

So stop arguing with Google. Four changes per page:

• H1 rewritten around the new keyword
• Meta description rewritten around it
• First 100 words rewritten so it appears naturally and early
• The body checked so the page genuinely answers that query

That last one is the one people skip. Swap the H1 and leave the article answering a different question, and you've made the page worse.

One check first. High impressions and low clicks has two causes with opposite fixes:

• Position 8 to 15: striking distance. Repointing moves it.
• Position 1 to 3 and clicks still falling: an AI Overview is probably answering the query above you. Rewriting the title won't fix that.

Using ChatSEO, list the pages getting impressions for keywords they
are not optimised for. For each page: the keyword I targeted, the
keyword Google ranks it for, current position, CTR. Then give me the
new H1, meta description and first 100 words.

Move 3: Regroup into silos

Every remaining page grouped by topic, a hub page per group, and the pages inside each group linked to each other. 40 pages in a pile became four or five clear clusters.

Slower to pay off. It compounds rather than spikes.

Cut first, because there is no point optimising a page you're going to delete. Repoint second, because those pages already have proven demand. Restructure last, on a clean set.

Never let a model make the irreversible call

Daniel Agrici built Jev SEO, an open source audit skill for Claude Code. In his own evaluation file, the model suggested removing substantial posts. Twice. Not thin pages. Real articles.

The fix was not a better prompt. It was a line of code: removal is blocked for any page of 600 words or more, whatever the model says.

-> Let the model rank. Let code, or you, hold the veto on deletes.

Module 10: If you're local, rank the profile, not the website

Local SEO is not regular SEO in a city.

When someone searches "plumber Houston", Google shows a map with three listings on top. Those are not websites. Those are Google Business Profiles. Fourth place is page two. Nobody scrolls the map.

Google picks those three on three factors:

• Proximity. How far the searcher is from your address. You cannot change it.
• Relevance. How sure Google is that you do this exact service. Content fills it.
• Authority. How much the web trusts you. Links fill it.

Most competitors filled in their profile once, three years ago. One category. Zero services. They ignore the only two factors they control.

Step 1: Measure before you touch anything

Rankings on maps change every two streets. Measuring from your desk tells you nothing. Use a grid rank tracker (Local Falcon or similar) on your primary keyword. You will see a green cluster around your shop that turns red three miles out. That border is what you are trying to push.

Step 2: Add every service Google suggests, not the ones you think of

In your profile, go to Services. Google shows pre-formatted services under your category, plus user-suggested ones in orange. Accept every orange one that applies. Then add custom services for every pattern from Module 2.

For each service, write a 300-word description mentioning the service, the city, and the problems you solve. Claude writes 30 of these in five minutes.

Step 3: Review velocity beats total count

A profile with 200 reviews and nothing this month loses to a profile with 40 reviews getting two a week. Ask every customer the day the job finishes. The single prompt that gets keywords into reviews:

Could you mention which service we did for you, and which neighborhood
you're in? It helps other neighbors find us.

Google reads those words. A review that says "great plumber in Cedar Park for our water heater" is worth three that say "great service".

Step 4: The website exists to feed the profile, not the other way around

On a local site, build one page per city or neighborhood you serve. Link your profile to the primary location page, not the homepage, if the address matches that city.

Write each place page about the place, not about you. The housing, the problems common there, the route your team drives. A page that only swaps the name is the doorway pattern from Module 8.

Part 3: Earn the authority

A perfect page on a weak domain ranks 40th. Nothing in Part 2 fixes that.

Authority is links from domains Google already trusts. You cannot prompt your way around them.

Five tiers, from free to paid. Run them in this order.

Module 11: Claim the links you already earned

The easiest links are the ones people meant to give you.

Move 1: Unlinked brand mentions

People wrote your name and forgot the link.

Search Google for "your brand" -site:yourdomain.com. Every result is a page that already knows you exist. Email the author, thank them for the mention, and ask if they can turn it into a link to the relevant page.

Conversion rate on these is roughly one in three. No outreach on cold domains comes close.

Move 2: Reclaim dead competitor links

Tools in your niche die, rebrand or get acquired. Every link to their dead site is now a 404.

Find them in ChatSEO or Ahrefs: check referring domains of a defunct competitor, filter on broken pages. Email the linking site: "Notice that resource went offline, we maintain an active equivalent here if helpful for your readers."

Module 12: Ask for the links nobody sells

The best links come from pages that don't have a "write for us" page.

When someone searches "best SEO tool for ecommerce", Google shows seven or eight roundups. Those authors update them once a year. Getting added to an existing ranking is faster than earning a new link from scratch, and it brings referral traffic the same afternoon.

The email that works is three sentences:

Read your roundup on [topic], saw [competitor A] and [competitor B].
We just launched [differentiating feature] on [our tool] which solves
[specific problem their readers have].
Would it make sense to include it in your next update? Here is a test
account if you'd like to try it: [link].

Antoine's rule: never offer money in the first email. The sites that take money are the sites that sell to everyone. Their links lose value every month. The sites that say yes because your tool is good are the ones Google actually counts.

The floor outcome is a mention without a link. Not a waste. Those roundups are what ChatGPT reads when someone asks which tool to use. Part 4 is built on that.

Module 13: Trade links that make both pages better

A pure link exchange (you link to me, I link to you) leaves a footprint Google sees immediately. A three-way exchange or a content collaboration does not.

When Axel audits link profiles, he looks for reciprocity ratio. If 40% of the sites linking to you are sites you link back to, you are running an exchange network and Google knows it.

The exchange that works is asymmetric:

• You publish a ranking of tools in an adjacent category (for example, we rank analytics tools, they rank SEO tools).
• You include them legitimately because they are good.
• They include you in their existing ranking of your category.

Both pages genuinely serve their readers. Both links sit inside relevant editorial text. The footprint is invisible because the categories differ.

The golden test: would you add this link if they never linked back? If the answer is no, don't do the trade.

Module 14: Earn links with something worth citing

The only links that compound are the ones you never asked for.

Journalists, bloggers and consultants link to things that save them time or make them look credible. Three formats earn 80% of passive links:

• Free tools. A character counter, a SERP previewer, an hreflang generator. They cost two days of dev time and rank on their own terms. People link to tools because you cannot paste a tool into your own blog post.
• Proprietary data. "We analysed 100,000 pages and found X". Every journalist writing about your industry needs a stat to back up their premise. If you have the stat, you get the citation. ChatSEO's link decay numbers in Module 15 are an example.
• Original definitions or frameworks. Name a problem your industry feels but hasn't labelled. The "DIFF GATE" in Module 7 is our attempt. When people reference the idea, they link the source.

Build one citable asset every quarter. It pays off on year two.

Module 15: Buy last, and read the footprint

Buying links is against Google's guidelines. Everyone does it. The difference between sites that survive and sites that lose their traffic in an update is whether the purchase looks like a purchase.

Before buying any link, run the seller's domain through these checks:

• Traffic trajectory: is the domain gaining organic traffic or losing it? A site losing traffic is shedding Google's trust. A link from it transfers less value every month.
• Outbound link ratio: does the site link out to 500 domains but only get links from 200? That is a link farm.
• Anchor text distribution: if 80% of incoming links use commercial anchors ("best CRM", "cheap shoes"), the site was built to sell links.
• Topical consistency: does a blog about gardening suddenly publish an article about crypto mortgages? Google's topic embeddings detect this instantly.

The silent link decay reality

Here is the number nobody in the link industry wants to talk about.

Over its lifetime, ChatSEO earned roughly 4,000 backlinks. Today, about 1,400 are still active and indexable.

Roughly two thirds of every link we ever earned was already gone without anyone doing anything wrong. Sites redesign, pages get deleted, blogs get shut down, authors remove references, platforms migrate CMS.

If you price your link campaigns assuming a link lasts forever, your economics are broken. Price against a 1-year survival rate. That discipline changes which links you buy and how much you pay.

Part 4: Get named where Google and AI look

GEO (Generative Engine Optimisation) is not a new discipline. It is classic SEO viewed through how LLMs retrieve information.

When ChatGPT, Perplexity or Claude need to answer a question they aren't confident about, they don't hallucinate a URL. They search Google, Bing or their internal web index. They read the top results. They summarise them.

If you are not in those results, you do not exist to the model.

Module 16: Target the query ChatGPT types, not the prompt your customer types

A user typing at 11 PM into ChatGPT writes a story: "I hurt my back at judo a month ago, the doctor said rest, but I wake up stiff every morning. What mattress should I look at?"

ChatGPT does not search Google for that paragraph. It breaks it down into fan-out queries:

• "best mattress back pain 2026"
• "mattress firmness lumbar spine support"
• "memory foam vs hybrid back pain"

Those fan-out queries look remarkably like classic commercial SEO keywords. Because that is what they are.

Jacky analysed over 5 million prompts and tracked what ChatGPT actually searches in the background. Three takeaways:

• The model rotates sources: across identical queries, it pulls from a pool of roughly 50 authority domains, citing 10 to 15 per answer.
• Topical queries fan out to year-stamped listicles: "best [tool] [current year]".
• Brand queries fan out to reviews and alternatives: "[brand] reviews", "alternative to [competitor]".

---
name: geo-queries
description: Build the list of queries AI assistants search for this site. Use when asked about GEO, ChatGPT visibility or AI citations.
---
1. Read keywords.md (from money-keywords).
2. For the top 10 commercial keywords, generate:
   - "best <keyword> <current year>"
   - "best <keyword> <current year> <country>"
   - the same two in English if the site is not English
   - "<brand> reviews", "<brand> alternative" for us and our top 3 competitors
3. Send ChatSEO, one query at a time, word for word:
   "Who ranks top 10 on <query>? Which of those pages are rankings, listicles or review pages, and does my site appear in them?"
4. Write geo-queries.md: query, ranking pages in the top 10, whether we are listed, our position, the domain behind it.
5. Never invent a volume or a ranking. Blank beats plausible.

Step 4 becomes the prospect list for the rest of Part 4. Every listicle ranking on a fan-out query is a page ChatGPT is likely to read.

What not to waste a week on

Google's documentation on AI features is blunt: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." You don't need new machine-readable files, AI text files or markup to appear there.

So if your agent proposes an llms.txt sprint, say no. The levers are the boring ones. Be cited by the pages the engines already cite. Keep the facts on your own pages current, in text, crawlable.

Module 17: Stop ranking yourself first

Putting yourself 1st in your own ranking was the cheapest GEO trick there was.

Paul did it for years. "Best SEO consultants in Paris", with Paul at number one, on his own site. ChatGPT cited the page, and cited him.

Then the source changed.

He was still in the answer. The page cited was now someone else's ranking.

Everyone ranks themselves first. The model has learned that, and discounts it.

Paul had Claude rewrite every ranking where he put himself first. His consultant site, its .com twin, and the ChatSEO blog, where we had ranked our own tool first. It burned through a day's worth of Claude usage in a few hours.

This is a trade-off, not a rule:

• For classic Google, a self-ranked page can still win. Paul still sees rankings like his at the very top of results.
• For GEO, it no longer gets you named.
• For your reader, it costs trust.

Paul removed his. If your rankings bring in clients from Google, test before you rewrite all of them.

---
name: self-rank-audit
description: Find and fix pages where we rank ourselves first.
---
1. List every page whose title or H1 contains "best", "top", "ranking" or a year.
2. For each: who is listed 1st, whether it is us, and the page's clicks over 90 days (from ChatSEO).
3. We are 1st AND the page gets clicks: do not touch. List them. I decide.
4. We are 1st and the page gets no clicks: rewrite the ranking on the merits. We may stay in the list, honestly placed.
5. Diff gate from ship-page. One page per run.

Step 3 protects your revenue. Never trade a page that sells for a citation you might get.

Module 18: Get ranked by others

A ranking on someone else's site is worth more than ten on yours.

A restaurant saying it is the best convinces nobody. A Paris guide naming it first does. Same for the model.

The ranking swap

Paul's rankings of agencies get him email. Agencies want in. He asks for no money. He asks for one thing back: a ranking on their own site that lists him, or ChatSEO.

Convertix is the live example. They published best SEO consultant in France, best SEO MCP. Paul published best Google Ads agency, best Meta Ads agency.

When Paul tested it live on a podcast, logged out, ChatGPT named him among the best SEO consultants in France. The source in the citations was Convertix's ranking.

Two rules make swaps compound:

• Recency. Refresh every year, with the new year in the title.
• Volume. Antoine thinks most businesses have at least 20 queries worth a swap.
• Your half of the swap is a page. Run it through page-brief and ship-page like any other. We never appear in a ranking of our own category.

Module 19: Find the Reddit threads the model is reading

Reddit is the single biggest source of citations in ChatGPT and Google AI Overviews for product recommendations.

Google signed a 60 million dollar deal with Reddit to access its data in real time. When an AI looks for authentic human opinions, it looks at Reddit.

How to find the threads that matter:

• Run your fan-out queries from Module 16 with site:reddit.com on Google.
• Look at the top 3 Reddit threads that rank on Google for those queries. Those are the exact threads the LLM scrapes.
• Contribute genuinely. Answer the question with substance. If your tool is relevant, mention it with full disclosure.

A single thoughtful comment in a high-ranking Reddit thread can drive citations across AI engines for months.

Module 20: Your domain gets 2 slots. Your brand can take 6.

Google has domain diversity rules: a single domain rarely gets more than two organic slots on page one.

If you only optimise your own website, your brand's presence is capped at two positions. But your brand is not limited to your domain.

A full SERP takeover targets six distinct surfaces:

1. Your own website pages: landing page or comparison page (max 2 slots).
2. YouTube review video: Google frequently displays video carousels on commercial queries.
3. Shorts or TikTok: increasingly indexed in mobile SERP packs.
4. Reddit discussion thread: ranks in Google's discussions and forums module.
5. Third-party listicle: a swap partner ranking in the top 5 mentioning your product.
6. Review platforms: Trustpilot, G2 or Google Business Profile ratings box.

-> Don't just try to rank the page. Occupy the SERP. When a customer searches your category, they should see your brand everywhere they look.

Part 5: Run it every week, and never faster

SEO is the same, slower.

Google deliberately delays the effect of changes. You cannot reverse-engineer it in a week.

The loop, the waits and the rules

The waiting rules you must memorize:

• After a change: a dip is normal. Give it up to two months. Reverting in a panic is how you teach Google you're testing it.
• Before buying links: three months on a page you know is good.
• On a new domain: six months before anything feels like it's working. Then a pivot, where new pages rank top 5 on arrival. That's when you turn the drip up.

The Monday review:

---
name: weekly-seo
description: Weekly SEO review from Search Console. Use every Monday, or when asked how SEO is going.
---
1. Send ChatSEO (siteId), word for word:
   "Compare my last 28 days to the previous 28. Which pages gained impressions, which lost clicks, and which keywords entered the top 10?"
2. Pages that lost clicks: check whether the ranking dropped or the CTR dropped. A lost ranking needs a better page. A lost CTR at the same ranking needs a better title, or means an AI answer now sits above us. Never the same fix for both.
3. Pages whose impressions jumped just entered the top 10. Queue them for page-brief. Top 10 to top 3 is the cheapest win there is.
4. Pages edited in the last 60 days: flag "wait". Do not touch them.
5. Pages edited 90+ days ago, still outside the top 3, with a good brief: candidates for links.
6. Write reviews/<date>.md and give me 5 actions, ranked by business value, not traffic.

Paul's own site now takes him about ten minutes a week.

The first Monday of the month, the GEO check. Rerun geo-queries, compare with last month, and check every third-party ranking you're listed in is still listing you. Then test 10 of your queries by hand, logged out, and note cited yes or no.

That last step stays manual on purpose. Jacky's team scrapes ChatGPT logged out because the API returns different sources than the web app. Ten queries by hand once a month is honest data. A dashboard built on the API is not the same thing.

Put both on a routine with /schedule in Claude Code and the reports wait for you.

The rules your agent runs under

Last, the rules your agent runs under, written before its first task:

ALLOW
- Read Search Console, analytics and ChatSEO
- Create drafts, write to the SEO log

REQUIRE MY APPROVAL
- Publish a page, or update a live one
- Change a title on a page that gets clicks
- Add or change a redirect

BLOCK
- Delete any page
- Edit robots.txt or canonical tags
- Send any email outside the company
- Buy anything

And every time you reject something, say why, and have the agent write the reason in a "SEO standards" file it reads before every draft.

-> That file becomes your editorial line, written by your own rejections. It outlives the tool.

Where I am wrong, or at least lucky

We started with an audience. Paul is one of the best known SEOs in France and has been doing it since 2021. Agencies email him to be in his rankings. Some of our links came because people already knew the name. A fresh domain with no brand starts slower than we did.

Most of the methods are not ours. Jacky, Antoine, Axel and Daniel are named where the method or the number is theirs. Jacky's numbers are French data on ChatGPT. Gemini and Claude may weigh things differently.

The local system is not our case study. ChatSEO is not a local business. Module 10 is what local agencies run for their clients, rebuilt so Claude does the writing.

We pick easy fights. Most of our patterns are low-competition queries. On a fresh domain going after commercial head terms, this exact process produces pages that rank 40th. The skills don't fix authority. They make sure you don't waste it.

It ages. The fan-out patterns already shifted once this year. Reddit's share of citations moved by a factor of six in a few months in Jacky's data. Some of Part 4 will be wrong by spring.

One company. Our blog, Paul's sites, a handful of customers. An anecdote, and I would rather say that than have it said in the replies.

The part I am confident generalises is the order. The part I am not confident generalises is the speed.

What I would do with a fresh domain tomorrow

• Connect Search Console to ChatSEO and run money-keywords before writing a single page
• Fix what already exists: cut, repoint, regroup
• Build one reference page per pattern by hand, with page-brief and ship-page
• Write the template and the process file, then publish 2 pages a week and never skip a week
• Claim every unlinked mention, then email 5 sites a day that rank where I want links
• Publish my own roundups, and trade placements once they rank
• Start geo-queries in month three, once there are pages worth citing
• Run weekly-seo every Monday and never touch a page edited in the last 60 days

In that order, because each one only works if the previous one exists.

Pick the money. Build better than the top 3. Earn the links. Get named by others. Wait. Repeat every week.

Outro & Author Notes

Thanks for reading!

🎁 If you want to hire your first AI SEO employee (for free) you can click here

Originally published by Nicholas DuLait (@NicholasDulait) on X (Twitter) on October 8, 2026. Official source: x.com/NicholasDulait/status/2108212094214529135.