YouMind
ログイン

How to use Claude Fable 5.2 for SEO so well it feels illegal

@NicholasDulait
英語2026年11月10日
1.3M
1.5K
150
69
6.7K

TL;DR

This article provides a comprehensive guide on using Claude Fable 5.2 for SEO, detailing workflows for keyword selection, content creation, and technical optimization.

FABLE 5.2 CAME OUT YESTERDAY AND THIS IS THE WHOLE SEO SETUP, FREE.

SO DON'T SCROLL PAST THIS ONE...

Anthropic shipped Fable 5.2 on November 9 at $8 per million input tokens, down from $10 for Fable 5.1, and up to 30% faster.

Everyone is going to test it on code this week, and almost nobody is going to point it at their own Search Console.

That's where the money is.

Nicholas Dulait - inline image

-> To get similar results for your business: Try CHATSEO.APP free for 7 days

This is the full setup. One keyword from your Search Console to one page Google has a reason to rank, with a prompt for every step.

The example all the way through is a furniture shop going after "oak coffee table". Swap in your own.

Save this. You'll come back to it.

Before we start: update Claude Code, or it won't take the model

Fable 5.2 is a new model id, and an old Claude Code doesn't know it exists.

So the first two commands are boring and you can't skip them:

text
1claude update
2claude --model claude-fable-5-2 --effort high

Fable 5.2 arrived in Claude Code 2.1.321. Run claude --version: anything older doesn't know the model.

Then remember what Fable 5.2 can't know on its own:

  • How many people search "oak coffee table" every month
  • Your clicks, impressions and positions
  • What Google shows on the results page today
  • Which competitor pages are taking your traffic

Ask it anyway and it answers from its training. Everything you build on that answer is a guess.

So the job splits in three:

  • Fable 5.2 is the hands. It plans, writes, checks.
  • ChatSEO is the brain. Through its MCP it brings your Search Console, keyword volumes, the live results page and your competitors' pages.
  • You are the boss. Nothing goes live without your yes.
Nicholas Dulait - inline image

To plug the brain in, create a ChatSEO account, add your site, connect Search Console, then add the MCP to Claude Code:

text
1claude mcp add --transport http chatseo https://api.chatseo.app/mcp

Now every prompt below works on real data, not on memory.

Nicholas Dulait - inline image

-> To get similar results for your business: Try CHATSEO.APP free for 7 days

1. Pick the keywords that pay before you write a word

A better model writing pages for the wrong keyword just gets you to zero faster.

"What is a coffee table" brings visitors, and nobody who types it is buying.

"Oak coffee table" is typed by someone with their card in hand.

You don't need a tool to read intent, because Google already tested it. Search the keyword and look at what ranks:

  • Blog posts, a definition box on top: informational. Skip it for now.
  • Product pages, category pages with prices: commercial. That's your target.

Then the second filter, the one that kills most projects: can you beat this results page at your size?

Nicholas Dulait - inline image
text
1Using ChatSEO, give me the best SEO strategy for my site. Then:
2
31. Keep only keywords where the current top 3 are product, category
4 or landing pages. Drop anything blog posts dominate.
52. For each keyword left, tell me if sites my size rank in the top 10.
6 If the top 10 are all big brands, mark it LATER.
73. Give me the 10 best keywords to start with: keyword, monthly
8 volume, the page type Google wants.
94. Use only the volumes ChatSEO returns. Never estimate one.

Rule 4 is the line people skip. If the model gives you a search volume no tool gave it, it's guessing.

The mistake that cost us the first months

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

The opportunity was real, with thousands of queries, low competition and obvious buyer intent.

So we launched fast and published fast.

It never ranked, not slowly, never.

We had a brand new domain, zero brand searches, zero backlinks, and a publishing rate that looked like every site Google files under "generated at volume".

-> We were nobody, showing up loud.

In March 2024 Google added scaled content abuse to its spam policies. It applies whether the content came from automation, humans, or both.

Volume is not the violation. Pages that add nothing are.

A faster model makes that mistake cheaper to repeat, which is exactly why you filter first.

2. Spend Fable on the page, not on the sorting

Fable 5.2 is the expensive engine, so the effort level is a budget decision.

Medium is the default in Claude Code. Anthropic says 5.2 uses fewer tokens at Medium than 5.1 did.

On Pro and standard Team seats, Fable is included for up to 50% of your weekly usage limits. Burn that on sorting keywords and you'll have nothing left for the page that matters.

Here's how I'd split the jobs:

  • Low: cleaning an export, grouping keywords, renaming columns.
  • Medium: reading the top 3 pages and pulling the outline.
  • High: writing the "oak coffee table" page itself.
  • Xhigh: the one page you'd be embarrassed to see on page four.
Nicholas Dulait - inline image

Run the sorting jobs in their own session, started at Low:

text
1claude --model claude-fable-5-2 --effort low

Already in a session? Type /effort and pick Low, no restart needed.

-> Same model, a fraction of the cost on the jobs that don't need the thinking.

3. One keyword, one page, better than the top 3

The most expensive mistake is writing a page you already have.

Two pages chasing "oak coffee table" split Google's attention, and neither wins.

Before writing, run the check Paul does by hand: Search Console, Performance, filter on the exact search, then the Pages tab. If a page already shows up, you improve it instead of writing a new one.

Then switch to High and build the page from what already ranks.

text
1Using ChatSEO, pull the top 3 results for "oak coffee table" and read
2their pages.
3
41. List every section all 3 have. My page needs each one.
52. List what only 1 of them has. Say if it is worth copying.
63. Find what none of them have that I could add, using only what I
7 can really show (photos of the grain, real dimensions, a care
8 guide, delivery times).
94. Write the outline: one H1 with the keyword, H2s, what goes in each.
105. Then write the page. Mark every fact you could not verify as
11 [CHECK].

Step 3 is the whole game. A page that only matches the top 3 is a tenth copy of the same page, and Google has nine already.

Nicholas Dulait - inline image

Anthropic says 5.2 communicates more clearly than 5.1. That makes the draft better, not finished.

Then you edit every line and every [CHECK]. Your name goes on the page, not the model's.

Nicholas Dulait - inline image

To get similar results for your business: Try CHATSEO.APP free for 7 days

4. Check what Fable wrote before Google does

Most pages that don't rank fail on something a checklist would have caught.

The list, for the "oak coffee table" page:

  • Title tag. The keyword near the start, under 60 characters, written for a click.
  • Meta description. One sentence that says what the reader gets.
  • One H1. With the keyword. Only one.
  • URL. Short, the keyword, no dates.
  • Internal links. At least 2 of your pages link to it, with words that describe it.
  • Images. Compressed, with alt text that says what's in them.
  • Structured data. Product, with price and availability.

Point Fable at the live URL and run this:

text
1Read the HTML of [URL], my page targeting "oak coffee table".
2
3Check it against this list and give me a pass/fail table:
4title tag, meta description, single H1 with the keyword, URL,
5internal links pointing in and out, image alt text, structured data.
6
7For every fail, write the exact fix: the new title, the new meta,
8the Product JSON-LD block ready to paste.

The JSON-LD is where the model earns its keep. Nobody writes structured data by hand twice.

The sitemap trap

Your sitemap tells Google which pages exist.

Paul 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, but the sitemap had stopped updating.

Nothing warns you. Check that your newest page is in your sitemap after every publish.

5. Every Monday, let Search Console tell Fable what to fix

After a few weeks, Google tells you what to do next, and most people never ask.

Look for searches where you sit between positions 5 and 15. Google already thinks you're relevant, so you're one good edit from page one.

text
1Using ChatSEO, look at my Search Console for the last 28 days.
2
31. List the searches where my average position is between 5 and 15,
4 with clicks, impressions and the page that ranks.
52. Group them by page.
63. For the 3 pages with the most impressions, tell me what to add or
7 change so the page answers those searches better.
84. Skip any page I edited in the last 60 days. Google hasn't
9 re-read it yet.

Run it at Medium. It's reading, not writing.

Nicholas Dulait - inline image

One hour a week. That's the whole loop.

Where I am wrong, or at least lucky

One day of Fable 5.2. The model came out yesterday. Nothing in this article is a ranking result from 5.2, it's the setup we already run with a newer engine in it.

The benchmarks are Anthropic's. Faster, cheaper, clearer writing: that's their launch post, not our test. Anthropic itself writes: "For most coding and knowledge tasks, we recommend that users start with Claude Opus 5.5, our strongest daily driver."

We started with an audience. Paul is one of the best known SEOs in France. Some of our early links came because people knew the name. A site with no brand starts slower than we did.

The part I am confident generalises is the order. The part I am not confident generalises is how much better 5.2 makes it.

What I would do tonight

  • Run claude update, then start Fable 5.2 at High
  • Plug the ChatSEO MCP so every answer comes from real data
  • Pick 10 keywords where the buyer has the card in hand and the top 10 isn't all big brands
  • Write one page at High, better than the top 3, and edit every line
  • Run the on-page check, then check the sitemap
  • Every Monday, fix the 3 pages closest to page one, at Medium

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

Update. Pick the money. Spend the effort on one page. Check it. Repeat every Monday.

Thanks for reading!

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

YouMindで再制作

Turn one viral article into a full content workflow

Collect the source, decode the pattern, create assets, draft the story, and distribute from one AI workspace.

Explore YouMind
クリエイターのために

あなたの Markdown をきれいな 𝕏 記事に

自分の長文を投稿するとき、画像・表・コードブロックを 𝕏 向けに整形するのは手間がかかります。YouMind は Markdown 全体を、そのまま投稿できるきれいな 𝕏 記事に変換します。

Markdown → 𝕏 を試す

解読すべきパターンをもっと

最近のバイラル記事

バイラル記事をもっと見る