The 12-step Claude SEO loop we used across 42 SaaS sites (1.3M organic clicks)
I kept seeing the same failure mode with AI SEO agents:
- They write every week, but traffic stays flat.
- The content reads like polished sludge.
- Backlinks, outreach, internal links, and old-page updates never happen.
So instead of asking Claude to “write an article,” we gave it the full SEO job across 42 SaaS projects. With this one SEO skill.
The combined result: 1.3M organic clicks.
These companies started at different stages. Some needed new content. Others already had traffic and needed links, distribution, or GSC-driven updates. The useful part wasn't a magic prompt. It was the loop.
1) Teach the agent how the SaaS makes money
We loaded the business before touching keywords:
- What the product does and who pays for it
- The problem it solves
- The money action: trial, demo, signup, or purchase
- Plans, prices, competitors, and existing pages
- Brand voice, writing rules, website, and publishing settings
This stops Claude from writing the same generic SaaS post for every company. A CRM for a five-person recruiting agency needs very different content from a CRM for a 2,000-person sales team.
2) Check the site and GSC before making more pages
A lot of SaaS sites already have pages Google almost likes.
We scanned for:
- Queries sitting around positions 5–20
- High impressions with weak CTR
- Pages losing traffic
- Old pricing or feature information
- Cannibalization
- Pages that need better internal links
A page at position 11 with 5,000 impressions may be a better job than another new blog post. The agent prepares a before/after change, the company reviews it, and the same URL gets updated.
3) Pick keywords that can bring traffic and signups
We used three buckets:
Buyer keywords: alternatives, X-vs-Y, pricing, integrations, free trials, and “software for [specific buyer].”
Problem keywords: how-to guides, templates, checklists, examples, workflows, and automation queries tied to the job the product solves.
Traffic/link keywords: statistics, original research, definitions, trends, benchmarks, and definitive guides.
For calendars with 30+ pages, we aimed for roughly:
- 30% traffic/link content
- 40% problem content
- 30% buyer content
But the buyer and comparison pages went live early.
4) Give every keyword the right page type
Claude chose the format before it wrote:
- Alternatives query → shortlist + real reasons to switch
- X-vs-Y → the same criteria applied to both products
- Pricing → current plans, limits, add-ons, and total-cost examples
- Integration → setup steps, screenshots, and honest limits
- Use case → the exact workflow for that buyer
- Template → something the reader can actually use
- Statistics → original data and clear sources
Sometimes the answer was “update the old page.” Sometimes the keyword was useless and got cut. Every keyword got a page type or got removed.
5) Give Claude proof so it stops writing AI slop
“One keyword + 2,000 words” is how you get a long article that says nothing.
Before writing, the agent received:
- Current prices, plan limits, and product docs
- Real screenshots and competitor facts
- GSC queries and customer questions
- Sales-call objections
- First-hand examples
- Honest product limitations
For comparisons, say where the competitor wins. For pricing, show what a real customer could pay. For how-to content, give steps someone can use today.
Proof is what makes generated content useful.
6) Connect every article and publish it
The agent finds useful pages already on the site and adds them to the brief. Every new page can link to the product, the right trial/demo, and relevant supporting articles without competing with another URL.
Then it schedules and publishes through WordPress, Webflow, Shopify, Ghost, Wix, Notion, GoHighLevel, Framer, or a webhook. Depending on the project settings, it stays a draft or goes live.
7) Put backlinks inside the same workflow
Good content still needs links.
Inside our backlink exchange, the agent finds relevant pages from real businesses and suggests one or two useful links where they actually help the reader.
Giving a useful link earns a credit. Credits help your site receive relevant links based on topical fit, a safe monthly pace, and which sites are a good match.
8) Pitch listicles that already rank
Say your buyer searches “best payroll software for startups.” A page-one listicle mentions three competitors and leaves you out.
The outreach system:
- Finds ranking listicles for buyer keywords
- Checks which ones mention competitors
- Removes weak/spammy pages
- Finds the author and verified contact details
- Drafts the first email, two follow-ups, and a LinkedIn message
The author still decides whether to add you. Some ask for money. You decide whether the page and price are worth it. One strong placement can send buyers while your own pages climb.
9) Publish useful versions outside your own site
One useful idea can become a native version for professional networks, Q&A sites, YouTube, and other third-party sites where buyers already spend time.
The goal is to put the brand in more places, earn useful mentions and links, and give Google and AI tools more credible pages that explain the product.
10) Track what ChatGPT and other AI assistants recommend
Google returns pages. AI assistants name products and show some of the sources that shaped the answer.
In one comparison of 10 Google buyer searches with 10 matching ChatGPT prompts:
- ChatGPT recommended 50 products
- Only 9 of those product domains appeared in the paired Google results
- Only 2 of ChatGPT's 10 top choices had their official domain in Google's captured results
- Only 5 of 20 visible ChatGPT source domains also appeared in Google for the same buying decision
We track buyer prompts across ChatGPT, Perplexity, Gemini, Google AI Overviews, and AI Mode. The gaps become content, outreach, or distribution jobs.
11) Let GSC tell Claude what to fix every week
Every week, scan for pages around positions 5–20, high impressions with weak clicks, traffic drops, missing sections, old details, and cannibalization.
The agent prepares a before/after update. The company approves it. Then the same URL gets improved.
12) Repeat what brings trials and demos
Clicks are an early signal. The business result may be a free trial, demo request, signup, paid account, or sales call.
Connect that result back to the keyword and page type:
- Comparisons bring demos? Build more useful comparisons.
- A stats page earns links? Use those links to strengthen buyer pages.
- A topic brings traffic and zero signups? Lower its priority.
The calendar gets smarter because it learns from what actually happened.
The weekly loop
Monday — find the best jobs
Scan GSC and AI mentions. Pick pages near page one and old pages worth updating.
Tuesday — research the next pages
Choose buyer/problem/link keywords, inspect what ranks, pick the format, and gather evidence.
Wednesday — write and review
Create new pages and old-page updates. Check facts and the before/after diff.
Thursday — publish and connect
Add internal links, useful exchange links, publish, and turn the article into native social content.
Friday — distribute and reach out
Pitch ranking listicles, follow up, prepare third-party versions, and record new links/AI mentions.
Then repeat.
If you want to try it with 10 pages
- Install the Distribb Skill in Claude, Codex, or ChatGPT.
- Connect your website and Google Search Console.
- Add your buyer, prices, competitors, and writing rules.
- Run the site/GSC audit.
- Choose 3 buyer, 4 problem, and 3 traffic/link keywords.
- Publish the three buyer pages first.
- Turn on the backlink exchange.
- Find the first 20 listicles ranking competitors.
- Track the 10 AI prompts closest to a demo or signup.
- Review GSC weekly and improve pages close to page one.
The big lesson for me: stop asking Claude for isolated articles. Give it a measurable growth loop.
I built this workflow into Distribb. You can try it free.
If you're using an agent for SEO, where does your loop break today—research, writing, publishing, links, or measurement?
