I Help Brands Get Recommended by AI and Found in Organic Search
Iceland
Joined October 2022
- Tweets1.3K
- Following353
- Followers3.9K
- Likes682
Agent SEO recipe #16:
Give Claude Code your crawl data:
- list every internal link that resolves through 2+ redirect hops
- resolve each chain to its final URL
- rewrite the ORIGINAL links to point there directly, then flatten the redirect rules themselves
Redirect chains are almost always one old migration nobody flattened. It's boring, mechanical work nobody ever gets around to which is exactly what agents are for.
If your agent does not have access to your CMS you can have it write it up as a task list, Google Sheet or use Maark.ai for it.
2
2
373
Agent SEO recipe #14:
Give Claude Code the DataForSEO MCP:
- pull People Also Ask for every keyword in the cluster (it's in the advanced SERP response — same price, just don't use the regular endpoint)
- dedupe the questions, group by subtopic
- questions your page skips become H2s; questions every competitor skips become your edge
The outline writes itself from questions people actually asked instead of having a writer guess at questions people don't actually have.
2
15
1,595
Target a topic on your site:
1. Buy a EMD for that topic
2. Add a bunch of content
3. Setup tracking
4. Google Search Console
After a few weeks you start to get super useful data around keywords you did not even know existed from the satelite site you can target on your main site.
186
One system for SEO and GEO means one place to track and report.
From there you create off site assets,
> buy more domains, lots of EMDs
> create more content and distribute to social media
> target keywords via parasite properties
> press releases for GEO
and it all becomes part of a large feedback loop...
Most brands think they need a second strategy for AI search and for me that sounds a bit silly....
You need one search strategy built to win on Google AND ChatGPT.
Here's the system 👇
1/ The work is the same in both:
- Content
- Site fixes (pricing, about, product pages)
- Off-site (outreach, mentions, reviews)
You don't have to do it twice if you organise your work correctly around what needs to happen but measure it in two places.
2/ Start with a hub: one area of your business you want to own.
A barrista show could look at Espresso machines. Coffee grinders. Pour-over gear.... each one is a topic HUB.
Google sees it as keywords. ChatGPT sees it as prompts. Same customer need.
3/ Then track two scoreboards:
SEO → rankings, clicks, revenue
GEO → mentions, citations, recommendations
4/ Each scoreboard shows different gaps.
Google shows the pages you're missing.
ChatGPT shows which pages get cited and which competitors get recommended.
Both feed one task list.
5/ Google clicks are shrinking. AI answers are growing.
The brands that win won't run two strategies. They'll run one system that shows up wherever customers search.
137
100% should be using keyword clustering to organise your keywords and not let Agents just run your campaigns out ther in the wild.
1
4
352
Most brands think they need a second strategy for AI search and for me that sounds a bit silly....
You need one search strategy built to win on Google AND ChatGPT.
Here's the system 👇
1/ The work is the same in both:
- Content
- Site fixes (pricing, about, product pages)
- Off-site (outreach, mentions, reviews)
You don't have to do it twice if you organise your work correctly around what needs to happen but measure it in two places.
2/ Start with a hub: one area of your business you want to own.
A barrista show could look at Espresso machines. Coffee grinders. Pour-over gear.... each one is a topic HUB.
Google sees it as keywords. ChatGPT sees it as prompts. Same customer need.
3/ Then track two scoreboards:
SEO → rankings, clicks, revenue
GEO → mentions, citations, recommendations
4/ Each scoreboard shows different gaps.
Google shows the pages you're missing.
ChatGPT shows which pages get cited and which competitors get recommended.
Both feed one task list.
5/ Google clicks are shrinking. AI answers are growing.
The brands that win won't run two strategies. They'll run one system that shows up wherever customers search.
Agent SEO recipe #12:
The AI-crawler firewall check, on a schedule instead of once:
- weekly: Claude Code requests your money pages as GPTBot, ClaudeBot, PerplexityBot
- challenge page or 403 = your firewall is eating AI crawlers (a spoofed user agent isn't a verified bot — confirm in your CDN's AI-crawler settings)
- alert on any change from last week
Firewall configs change without SEO in the room. The agent notices the day it breaks, not the quarter after.
2
1
142
Agent SEO recipe #10:
Give Claude Code your crawl's outbound-link list:
- request every external URL, follow the redirects
- redirect → rewrite your link to the final URL
- dead → replace with a live equivalent or cut the claim it supported; never blind-delete the sentence
This is the chore every site defers forever. An agent clears it in an afternoon and monitors and reruns it every month
1
4
279
Agent SEO recipe #9:
CTR fixes grounded in the SERP, not in taste:
- for every page with weak CTR at position ≤10, pull the top-10 titles
- have Claude Code extract the pattern that wins there: year? number? brand? question format?
- rewrite toward the pattern, keep the promise honest and have the agent re-read CTR at 28 days and revert the losers
If nothing else, front loading your main keywords in the meta title tends to improve CTR quite a lot for un-optimized pages.
Agent SEO recipe #8:
Stop labeling keyword intent by reading the words. Have Claude Code read the results:
- pull the top 10 for each keyword cluster
- count what actually ranks: product pages? categories? guides? comparison posts?
- 7 of 10 results are products = Google decided it's transactional, whatever the keyword looks like. Mostly guides = informational. Mixed = look closer before betting a page on it
A keyword tells you what someone typed. The SERP tells you what Google thinks they meant. Brief the page type that's already winning.
1
199
Agent SEO recipe #7:
Before you hand an agent a paid API, build the meter. Ours is one SQLite table and three rules:
- every call gets logged with provider, endpoint, cost, and what it was for
- the agent reads today's total BEFORE it calls, and stops at a daily cap ($5 is plenty while you're testing)
- one env var kills all spending without touching code
We learned this the honest way by accidentally draining all our accounts in a few hours without knowing it was happening.
An agent will happily make ten thousand calls doing exactly what you asked it to. The meter is what makes that safe.
228
Agent SEO recipe #6:
Stop briefing writers from vibes. Give Claude Code Firecrawl + DataForSEO and have it build the brief:
- top-10 SERP for the target: page types, title patterns, shared subtopics
- scrape the top 5, extract every entity and question they cover
- pull your own GSC queries for the page — unanswered ones become H2s
Works the same whether a person writes it or another agent does. Most bad content was already bad at the brief stage so this is an effective way to improve that.
2
5
427
Agent SEO recipe #5:
Give Claude Code your crawl export and an OpenAI key:
- have it embed the body text of every page (text-embedding-3-small costs $0.02 per million tokens — an entire site is a few cents)
- ask for page pairs that cover closely related topics but don't link to each other in either direction
- add the links where they make sense — in the body text, both ways
Most sites have years of content written by different people who never read each other's pages. The agent finds every unlinked pair in one run.
We do this on client sites and always walk away with a list of pages we can connect.
2
1
456
Agent SEO recipe #4:
Give Claude Code the Firecrawl MCP:
- firecrawl_map your domain — every URL it can discover
- diff that against your sitemap
- diff both against pages earning GSC impressions over 90 days
Three lists that never match.
Mapped-but-not-in-sitemap leans orphan.
In-sitemap-but-zero-impressions is your deadweight review.
The gaps between the lists ARE a list you can work on to improve your SEO performance, of course you can then layer on more data to do more work but that is good start.
Agent SEO recipe #3:
Give Claude Code the DataForSEO MCP and your GSC query export:
- pull the top 10 for every query (SERP API, standard queue)
- group keywords wherever 5+ of the 10 URLs match — that's ONE intent
- one page per group; merge the rest, 301 the losers
Google already ran the intent test on every keyword you care about.
8
15
2,025
Agent SEO recipe #2:
Want to know if AI assistants ever recommend you? Don't ask ChatGPT once and call it a day as the answers change run to run.
- write the 20 questions a buyer would actually ask based on actual search data ("best X for Y")
- have Claude Code ask each one ~10 times through the OpenAI and Perplexity APIs
- log every brand that comes up, count the mentions, re-run weekly
Now you have a measurement instead of a screenshot. We've run 7,048 of these probes across 5 engines and 97.7% every week.
2
6
399
Agent SEO recipe #1:
An SEO agent needs two things to be effective, so you can use them at a basic level.
a way to read pages, and access to search data. Two commands give Claude Code both:
- claude mcp add firecrawl -e FIRECRAWL_API_KEY=... -- npx -y firecrawl-mcp
- claude mcp add dataforseo -e DATAFORSEO_USERNAME=... -e DATAFORSEO_PASSWORD=... -- npx -y dataforseo-mcp-server
Now it can crawl any site and pull live SERPs, volumes, and rankings.
Things you can ask for today
which page types win the top 10 for your money keywords · who ranks for "best [your category]" · your competitor's most-linked pages · volumes for your whole GSC query export · a crawl of your own site with the issues grouped by template.
Setup is ten minutes. Firecrawl's free tier is 1,000 credits a month (one page = one credit); DataForSEO wants a $50 deposit, then SERPs run $0.60 per 1,000.
8
9
924
Easy SEO win #60:
- write your 20 money prompts ("best X for Y")
- ask ChatGPT + Perplexity each one ~10 times (once is a coin flip)
- log every answer that doesn't name you
We ran 7,048 of these probes across 5 engines: 97.7% named NO measured brand.
The shelf isn't taken. It's empty.
2
386
Easy SEO win #59:
- pull one day's clicks from GSC three ways: by property, by page, by query
- we measured a site: 92 vs 154 vs 26. same site, same day
- all three are "correct"
Report totals by property. Drill down in one view. Never compare across views.
1
6
272
Easy SEO win #58:
- pull your full URL list (sitemap or crawl)
- subtract every page GSC shows impressions for over 90 days
- what's left is your zero-impression set. merge, 301, or improve each one
Healthy sites we work on: ~2% of pages. We've measured a site at 52%.
Never blind-delete — a deleted page throws away every link and ranking it earned.
2
195