Three days of LLM SEO, in one hour.
Case studyEnd-to-end process automationVideo 3:10
We chained a three-day senior LLM SEO audit into one automated run for an SEO client, who isn’t named. It takes a business description, competitors and a site crawl, builds the keyword research, generates the queries people ask LLMs and runs a semantic audit of the site, then delivers FAQs, page briefs and recommendations in a Google Sheet. A run that took about three days of senior time now takes about an hour. It has run more than 25 audits, mostly on ecommerce sites, and was built in eight weeks.
- Client
- An SEO client, not named
- Category
- Workflows
- Stack
- Claude, Google Sheets, search APIs, crawl tools
- Time to value
- 8 weeks, build and tune
- Status
- Live, 25+ audits run, mostly ecommerce
- Updated
Results
Measured across more than 25 audits, mostly on ecommerce sites.
| Metric | Before | After | Measured over | How measured |
|---|---|---|---|---|
| Audit time | About 3 days of senior time | About 1 hour | 25+ audits | Run time per audit. |
| Audit quality | Previous manual reports | Better in output testing | 25+ audits | Output tested against the previous manual reports. |
| AEO rankings | Before the audit | Improved | Audited sites, mostly ecommerce | Ranking changes after the recommendations went live. |
See it in action
Watch the chain run from a raw crawl file to a tabbed Google Sheet of explicit LLM SEO recommendations, hands-off.
Read the transcript
From Vimeo’s auto-generated captions, lightly edited.
What it is
So the tool I’m going to show you here that we’ve built is ultimately LLM optimisation. So what it does is first we do a very… my background is head of SEO and CEO of an SEO agency, so we do very detailed SEO processes powered through AI.
Keyword research
So the first point here, and I won’t show you all because it takes about 15 minutes per step, but it runs in the background, is keyword research. So you go in, you upload a description of the business, who the competitors are, a crawl file, and it will then create 600 keywords that are better than a human can do.
LLM query generation
The next step is LLM query generation. So it takes those keywords and, using a very SEO-focused logic, identifies behavioural patterns, checks search volume for them to then pull and generate LLM queries.
Semantic analysis
And then the last one is… then, what it does is, so let’s pull LLM queries. Let’s just put in context: “On Running is an emerging trainer brand focusing on e-commerce opportunities around apparel and trainers.” I put in the website URL, I put in the number of prompts, and I run the semantic analysis. So what this does, it loads the prompts, it understands what the prompts are to then identify what the prompts should be, and then it goes and looks at the content of the website to then generate FAQ prompts if necessary. So it goes in detail, pulls competitive content to understand what people are saying around it, social forums, et cetera. Understands that semantically and then comes back to then feed into it. It does take a little while, so I’ll probably fast-forward through this bit as it generates for you.
The output
But then what happens at the end of it, it will showcase very detailed prompts. So what it’s done is it’s gone and semantically analysed all the content using crawl files, site maps, and everything else. And then what it’s done is it’s created FAQs. So it’s identified 39 FAQs where I can bring this, focusing particularly on Cloud because that’s their key product, then different genres that they want to target, like women’s, Cloudtec, et cetera. And then it’s gone and created briefs, and it’s created a detailed breakdown of briefs. And I can open the sheet here, and this will open the sheet. So in this, we have all of the FAQs with all the answers. So if we want to generate that and get that on their webpage, it’s there, whether it’s a new landing page, the intent, the categorisation of the keyword and query. And then it’s created a list of all of the prompts. So: what are the best trail running shoes for beginners who want to start hiking? It’s got proposed page structure, a short brief, and everything that they could give to their team.
The result
So what it’s done is essentially consolidated the end-to-end process of generating keywords, identifying what those queries are through LLMs, and then generating on-page optimisation automatically, done at a scale that was previously humanly impossible. So the whole process end to end takes about an hour in total with it running in the background, but the output you get is the same as a team taking a week to do.
What it does
Three phases, one run. Keyword research feeds query generation, and query generation feeds the semantic audit. Three days of senior LLM SEO work compressed into a single chained workflow.
12-step keyword discovery
Pulls from titles, URLs, headers, competitor pages, People Also Ask, AI search and social signals, combined into one prioritised list.
Five long-tail queries per keyword
Every approved keyword expanded into five LLM-search variants based on semantic behaviour.
Three-way scoring
Each query scored on LLM-query likelihood, relevancy and search volume signal. Top opportunities surface automatically.
A five-category audit
Final recommendations sorted into a tabbed Google Sheet: new content, update targeting, add FAQ, restructure, or other.
The workflow
Three phases that chain. Each phase’s output is the next phase’s input, and the only manual checkpoint is between the first two.
-
Keyword research
Combine business context, target areas and a fresh crawl file into a prioritised, categorised keyword list, ready for review. The user approves this phase before the chain continues. It is the only manual checkpoint in the whole workflow.
What runs inside this phase
- Starting inputs
- Crawl file, business context, target areas, competitor URLs.
- Flows to query generation
- An approved keyword Google Sheet.
- 12-step combineCrawl file, business context and target areas combined using a 12-step approach to seed the initial keyword universe.
- Multi-source pullKeywords pulled from titles, URL inclusion, header structure, competitor pages, People Also Ask, AI search behaviour and social.
- Prioritise and categoriseThe full list is deduplicated, scored and grouped into topical categories so the most valuable keywords surface first.
- User approvalThe filtered list is presented for review. Once approved, the system generates a shareable Google Sheet for phase two.
-
Query generation
Turn each approved keyword into five long-tail LLM queries, score each one on three signals, and map them back against the crawl file to see what your site already covers and what it doesn’t.
What runs inside this phase
- Flows from keyword research
- The approved keyword Google Sheet.
- Flows to the semantic audit
- An updated sheet with keywords, queries and prioritised LLM queries.
- Ingest keywordsThe approved keyword document is parsed and prepared for semantic expansion.
- Five long-tail variantsFive LLM-search variants generated per keyword from real semantic behaviour, not template substitution.
- Three-way scoringEach variant scored on LLM-query likelihood, topical relevancy and search volume signal.
- Map to the crawl fileFinal LLM queries mapped against your existing crawl to show what your site already covers and where the gaps are.
- Updated sheetAn updated Google Sheet with keywords, queries and a prioritised LLM query list, ready for audit.
-
Semantic audit and recommendations
Take the full set of scored LLM queries, compare every one against the site mapping, and sort opportunities into five action categories. Each category gets its own tab in the final sheet, with explicit recommendations.
What runs inside this phase
- Flows from query generation
- The full set of scored, mapped LLM queries.
- Final deliverable
- A tabbed recommendations sheet with explicit guidance per opportunity.
- Ingest queriesThe full LLM query set is loaded into the audit engine with all mapping context.
- Compare to the mappingEach query checked against the crawl-to-query mapping for coverage strength and content fit.
- Sort into five categoriesNew content required, update targeting, add FAQ, restructure on-page, or other.
- Tabbed recommendationsEach category gets its own tab in the Google Sheet, with named recommendations and the reasoning behind each one.
The problem
LLM optimisation was three days of senior expertise. Each phase was its own engagement, mistakes cascaded downstream, and junior staff were locked out.
LLM SEO is genuinely hard. Keyword research alone takes a day if you do it properly: pulling from competitors, People Also Ask, AI search behaviour, social signals and your own crawl, then prioritising the mess into something usable. Query generation takes another day, turning every keyword into the variants real users and LLMs actually search. The semantic audit, comparing queries against your site map, sorting opportunities and writing explicit recommendations, takes a third.
It scaled badly. Each phase was its own engagement and a junior couldn’t run any of it alone. A small mistake in phase one would silently compound through phases two and three, only surfacing at the end of three days.
By day three, no one wanted to look at the audit again, even when it was right.
- 3 days of senior time for a single LLM optimisation audit.
- Three separate engagements stitched together by hand, each with its own quality risk.
- A phase-one miss became a phase-three problem.
- Junior staff couldn’t deliver on their own, capping how many audits the team could run.
The solution
A single automated chain. Three phases, each with its quality gate built in, each handing its output directly to the next.
One chain, three phases, an hour instead of three days
The chain runs keyword research, query generation and the semantic audit back to back. Phase one’s approved keyword sheet feeds straight into phase two, and phase two’s scored queries feed straight into phase three. The output is a tabbed Google Sheet with explicit recommendations in five action categories. The user approves once, between phases one and two; the rest runs hands-off.
- Multi-source keyword engineCrawl, context, competitor, People Also Ask, AI and social signals combined into one prioritised list using a 12-step approach.
- Semantic query expansionFive long-tail LLM variants per keyword, scored on likelihood, relevancy and search volume signal.
- Tabbed audit outputEvery opportunity sorted into one of five action categories, each with its own tab and explicit recommendations.
- Keyword research412 keywordsApproved
- Query generation2,060 queriesScored
- Semantic audit134 actionsDone
| Sheet tab | Actions | Share |
|---|---|---|
| Update targeting | 54 | |
| New content | 38 | |
| Add FAQ | 21 | |
| Restructure | 12 | |
| Other | 9 |
Want to see it?
Arrange a call and we will walk you through this chain live, plus the adjacent workflows we could build for your team.
Arrange a callThe outcome
Three days became about an hour. Across more than 25 audits, mostly on ecommerce sites, the output tested better than the previous manual reports, and AEO rankings improved.
- 1hrPer auditDown from about three days of senior time.
- 25+Audits runMostly on ecommerce sites.
- 8wksTo build and tuneFrom kick-off to running live audits.
Questions
How long does an automated LLM SEO audit take?
About an hour, running in the background, compared with about three days of senior time by hand.
How many audits has it run?
More than 25, mostly on ecommerce sites.
How do you know the audits are better?
We tested its output against the previous manual reports, and AEO rankings improved on the sites audited.
How long did it take to build?
Eight weeks, including tuning.
More case studies
All our work-

Next Gen Intelligence: raw signal to brief-ready concepts
Case studyIntelligenceVideo 2:40
60,000 qualitative documents turned into trends, platforms, personas and cited creative concepts.
-

Scope Assistant: a scope in two minutes
Case studyOperationsVideo 1:26
Reads a client call, runs a 22-prompt framework and lands a formatted scope in Drive.
-
Weekly check-inClient callSend the revised media planBook the creative reviewRecap and next steps
Team Agent Assistants: the admin already handled
Case studyOperations
Reads each meeting transcript, reconciles tasks in Trello and drafts the client follow-up in your voice.
Your three-day audit, in an hour
Get in touch and we will walk you through the chain live, plus the adjacent LLM workflows we could build for your team.