The cycle

Five phases. Ninety days. Then again.

Phases overlap on purpose: we don't wait for a perfect audit to start fixing obvious problems, and we don't stop measuring while we optimize.

What we need from you

  • Around two hours of subject-matter-expert time per month
  • Access to your CMS, analytics and (ideally) server logs
  • One point of contact and a weekly 30-minute check-in
  • Sign-off on content and outreach within five working days
01
Weeks 1–3

Audit

We run 150–400 priority prompts through six engines, record every citation and claim, benchmark competitors and diagnose the gap: content, entity, authority or access.

  • Citation and share-of-voice baseline
  • “What AI says about you” accuracy review
  • Technical access and extractability scores
02
Weeks 2–4

Prompt map

We expand the prompt set into a full universe — the questions your buyers actually ask, in the words they use — clustered by intent and stage, and prioritised by value and winnability.

  • Interviews with sales and support
  • Conversation-log and query mining
  • Value × winnability prioritisation
03
Weeks 4–10

Optimize

Technical fixes ship first (they're fast). Then answer-ready rewrites of the pages that matter, new pages for prompt gaps, structured data and entity clean-up across the web.

  • Crawler policy, llms.txt, rendering
  • Extractability rewrites and new answer pages
  • Schema and profile synchronisation
04
Weeks 5–12

Amplify

We work the source map: the publications, review platforms, communities and comparison pages your engines already cite, earning your brand a genuine presence in each.

  • Data-led PR and expert commentary
  • Review-platform and community programme
  • Comparison and “best of” inclusion
05
Weekly, always

Measure

Every prompt, every engine, every week. We report what moved, why we think it moved, and what we're doing next — then roll into the next 90-day cycle with a sharper map.

  • Weekly digest and live dashboard
  • Monthly narrative report
  • Quarterly strategy review and re-plan
How we measure

Five numbers that tell you if it's working.

Traditional analytics can't see AI answers. So we built a monitoring platform that runs your prompts on a schedule and scores the results.

Citation rate

The share of your priority prompts where at least one engine cites you. The headline number.

Share of voice

Your citations as a share of all citations across the prompt set — versus each named competitor.

Citation position

First source or fifth? Position predicts clicks and how much of the answer is about you.

Sentiment & accuracy

What the answer says about you — and whether it's true. Wrong pricing in ChatGPT costs real deals.

AI referrals & pipeline

Sessions, sign-ups and revenue from AI-assistant referrers, tracked in your analytics and CRM.

Different game

How LLM search optimization differs from SEO.

Good SEO is a head start, not a substitute. The unit of competition changed, and so did most of the signals.

DimensionTraditional SEOLLM search optimization
Unit of competitionA rank on a results pageA citation inside one synthesized answer
What winsRelevance, links, page experienceExtractability, corroboration, entity clarity
Content shapeLong, keyword-targeted pagesDirect answers, definitions, tables, original data
Authority signalsBacklinksMentions across the source types engines trust
MeasurementRankings, impressions, clicksCitation rate, share of voice, position, sentiment
Feedback loopWeeks to monthsDays to weeks for retrieval-based engines
Zero-clickSometimesBy default — brand recall matters as much as traffic
Where you show upgoogle.comChatGPT, Perplexity, Claude, Gemini, AI Overviews, Copilot…
Ground rules

How we work — and what we won't do.

01

Humans first

Everything we write is written for people. Models reward that; readers demand it. We never ship content we'd be embarrassed to put a name on.

02

Total transparency

You see the same dashboard we do, updated weekly. Every citation, every prompt, every engine — no cherry-picked screenshots.

03

No dark patterns

No fake reviews, no astroturfed threads, no hidden-text or prompt-injection tricks. They stop working the moment models notice — and they always notice.

04

Engine agnostic

We don't bet on one assistant winning. We report per engine and build for the underlying behaviours they share.

Questions

About the process.

Why 90 days?

It's long enough for content and outreach to be picked up by retrieval systems and reflected in answers, and short enough to keep the team honest. Most clients see first movement in four to eight weeks; the 90-day mark is where we re-plan with real data.

How many prompts do you track?

Starter tracks 150, Growth 400 and Enterprise 1,000+, each run weekly (daily on Enterprise) across six engines. We'd rather track the right 150 than a noisy 5,000 — the prompt-map phase is about choosing well.

Do you use AI to write the content?

We use it the way any good writer does now: for research, outlines and first passes. Everything that ships is written or rewritten and fact-checked by a human editor, and reviewed by your subject-matter experts. Original data and real expertise are the things models can't get anywhere else — that's where the human time goes.

What happens when an engine changes how it cites?

We see it in the monitoring within days, because we run the same prompts on a schedule. We tell you what changed and what we're adjusting. Because we build for shared behaviours — extractability, corroboration, entity clarity — engine changes usually shift the numbers, not the strategy.

See the process applied to you.

Book a call and we'll run a handful of your prompts live — you'll see what the engines say about you today.