One industry-agnostic, phase-gated system. SEO and GEO run in parallel every week, measured against a locked prompt set and proven against the baseline - not a vague promise, the exact operating standard we run on every engagement.
6 phases24 weeks1,400+ tracked tasks27-sheet execution workbookReach the page. Render the meaning. Earn the trust. Measure the truth.
Non-negotiable
Three principles that override everything.
They protect the client and the agency, and they are what makes the work survive model updates and platform crackdowns.
Principle 1
llms.txt is hygiene, not a growth lever.
The file is llms.txt / llms-full.txt (lowercase, with the 's'). We ship it for agent and developer routing, never sold as a citation driver. As of 2026, answer engines fetch it rarely and Google has declined to support it.
Principle 2
Off-site authority is earned, never manufactured.
No fake reviews, no planted or programmatic “unlinked mentions.” The FTC fake-review rule and platform terms make manufactured consensus a legal and reputational liability, and engines increasingly detect it. We deserve the consensus; we do not fake it.
Principle 3
Every number is sourced or struck.
No unattributed statistics - including the well-known “+30–40% citation” figure, which comes from one synthetic benchmark and does not transfer to commercial sites unverified. We cite the primary source with its limits, or reframe the claim as a hypothesis to test against the client's own baseline.
Also enforced on every report
AI Citation ≠ AI Referral ≠ AI Conversion. Being named in an answer, receiving a click from an AI surface, and completing a purchase are three different events, tracked separately.
Set-Up
The measurement instruments, built per client.
Before Week 1 closes, two instruments are built and version-controlled. Everything downstream is measured against them, and neither is changed without a logged rationale. If either drifts, every comparison becomes noise.
The core keyword universe
The finite set of commercial terms the client must rank for and be cited on. Grouped into 4–6 clusters (a product/service line, a category, or a use-case), each keyword tagged by search intent, cannibalization-risk pairs flagged. One canonical URL per keyword.
The diagnostic prompt set (~15–25 prompts per vertical)
The real conversational questions buyers ask ChatGPT, Perplexity, Gemini and Claude when entering the category - the AI-visibility test set. Run on a fixed cadence to compute Share of Voice, citation rate and competitor displacement.
Prompt taxonomy
Prompt type
Engine behaviour
Primary GEO lever
Navigational
Returns a direct link; brand must own the result
Organization schema + indexing
Informational
Synthesizes a multi-source explainer with citations
Model-specific tuning and the highest-intent comparison prompts.
Month 6 · Wk 21–24
Trust & consolidation
Attribution, automation, projections and handover.
Core monthly production cadence
4–6 content assets built around the month's cluster and diagnostic prompts.
FAQ section added or expanded on every priority page (FAQPage schema as extraction hygiene).
1 keyword & prompt-set rotation per the cluster schedule.
2–3 quality guest posts on relevant, authoritative publications - no link farms.
1 authority/data article (e.g. a cost, benchmark, or trend piece the brand can own).
1 internal-linking sprint + 1 backlink-gap and unlinked-mention review.
2–4 digital-PR pitches to relevant media and industry outlets.
Technical SEO monitoring + AI-visibility measurement across ChatGPT, Perplexity, Gemini and AI Overviews.
Content freshness / factual-accuracy check on published pages (update only on genuine change).
Governance Layer
Gates, cadence and change control.
Measure before you touch, ship reversible changes, and prove every delta against the baseline. This is what separates a professional agency from an incident waiting to happen.
Phase-End Go/No-Go Gates
Gate
Week
Pass condition
Phase 1
Wk 4
All target AI + search agents return 200 and are logged; SEO + SoV baselines locked and timestamped.
Phase 2
Wk 8
Schema validates clean; entity consistent across controlled surfaces; first-chunk density confirmed.
Phase 3
Wk 12
Top pages re-engineered and extraction-tested; Trust Triad + author signals live.
A one-page performance scorecard is produced at the end of every even week (Wk 2, 4, 6 … 24). Each reports the same six lines so trends are comparable across the engagement: SoV delta (vs last fortnight + vs baseline), source-inclusion change, AI-referral sessions, deliverables shipped vs plan, displacement wins, and risks + next-fortnight plan.
Store any unavoidable shared secret only in the team password manager - never email, Slack, or docs.
Request least-privilege roles; no “admin” where “editor” suffices; enable 2FA.
Log every grant in the client access register; revoke every credential within 24h of engagement end.
Risk register & rollback triggers
The two highest-risk activities are the Week-3 WAF/crawler-access change and the Week-6 schema deploy - both carry explicit, monitored rollback triggers.
Risk
Rollback trigger
Action
Owner
WAF change blocks human traffic
Bot spike + conversion drop >5% within 48h
Revert rules; re-test agents; notify client <4h
Client infra / Specialist
Schema deploy breaks rich results
GSC coverage-error spike within 72h
Revert JSON-LD; reconsideration; root-cause first
Specialist / Lead
llms.txt misrepresents brand
Client flags a factual error in an AI description
Update immediately; note which models cached old version
Six phases. Twenty-four weeks. Every week, in full.
Objective, SEO protocol, GEO protocol, deliverable and definition of done - the same structure, every week, so nothing is a black box.
Phase 1 · Weeks 1–4
Technical Foundation, Baselines & Access
Establish access, capture true pre-work baselines, and make the site machine-legible to search crawlers and AI agents before any optimization ships. The whole pipeline is gated here: if a crawler cannot reach a rendered 200-status page, nothing downstream matters.
Go/No-Go: Every target AI + search agent returns 200 and is logged; SEO and SoV baselines are locked and timestamped. Rollback: any WAF change harming human traffic is reverted per the Risk Register.
Phase 2 · Weeks 5–8
Site Architecture, Entity SEO & Schema
Turn the keyword universe into an information architecture, resolve cannibalization, and build a connected entity graph the engines can resolve - starting with the highest-value cluster.
Go/No-Go: Schema validates clean, entity is consistent across controlled surfaces, and first-chunk density is confirmed on money pages. Rollback: rich-results break in GSC → revert JSON-LD.
Phase 3 · Weeks 9–12
Content Production, On-Page SEO & AEO
Launch the priority content cluster and make the highest-value pages answer-first and citation-ready: compliant definitions, native tables, question-shaped headings, and a full FAQ layer.
Go/No-Go: Top pages re-engineered and extraction-tested; Trust Triad and author signals live.
Phase 4 · Weeks 13–16
Backlinks, Digital PR & External Authority (Earned)
Close the backlink gap against competitors, launch the next cluster, and build durable earned presence. Everything here is earned and disclosed - no fake reviews, no planted mentions, no manufactured consensus.
Go/No-Go: First earned citations landing and directories consistent; review flow compliant and live.
Phase 5 · Weeks 17–20
SEO + GEO Performance Optimization
Let data - not assumptions about how any one AI model works - drive the next round of optimization. Launch the next cluster and comparison content, tune to the dominant engine, and prove movement against the Week-4 baseline.
Go/No-Go: Positive SoV delta vs the Week-4 baseline on the locked prompt set.
Phase 6 · Weeks 21–24
Attribution, Trust, Automation & Handover
Close the measurement loop, separate citation from referral from conversion, automate monitoring, model the future honestly, and hand over a durable, repeatable operating system.
Every metric has a precise definition, a fixed source, and a documented collection method so results are reproducible. The baseline is always visible on every chart - a number without its baseline is meaningless.
Metric
Source
Definition
Citation Share of Voice
GEO Metrics / Otterly
% of the fixed prompt set where the brand appears, tracked vs competitors over time.
Source Inclusion Rate
Profound / Keytomic
Of those appearances, how often an active link to the domain is attached (unlinked vs linked).
AI Recommendation Rate
Manual / tool API
% of transactional prompts where the brand is explicitly recommended (ideally top-3).
AI Referral Traffic
GA4
Sessions from AI referrers via the full custom channel group + regex (SOP K2).
Brand Search Lift
GSC
Growth in exact-match brand-query impressions - the downstream echo of being cited.
Projection discipline
Citation-velocity curve: citation/entity acquisition vs optimization output (pages shipped, schema connected, earned references) - show the fitted curve and its assumptions.
Referral-traffic value: sessions × niche Google Ads CPC, labelled a paid-equivalent proxy - never booked revenue, never guaranteed bookings.
Always present three scenarios (Conservative / Base / Optimistic) with the assumptions that separate them. Never a single-point projection.
QA Gate
The master practice checklist before anything ships.
The consolidated pass/fail view across twelve SOP domains - the engagement-wide audit and the QA gate before anything reaches a client's live site.
Ref
Standard met
A1
Robots.txt clean; per-agent policy explicit; training posture decided; sitemap referenced
A2
All target AI agents return 200 at the edge; no JS-challenge; client-infra signed off
A3
Server logs confirm real bot hits on priority URLs; no 4xx/5xx to intended bots
A4
Sitemaps list only canonical 200 URLs; IndexNow live for Bing/Yandex
B1
Core content + headings + links + JSON-LD present in raw HTML (SSR/SSG verified)
B2
Money pages 200; no redirect chains; canonicals correct; duplication handled
C1–C3
DOM simplified; Markdown mirror clean; core answer within ~512 tokens (retrievability tested)
D1–D4
Connected @graph, accurate + server-rendered, zero validator errors; 72h GSC watch clean
D2/D3
sameAs to live authority nodes; Wikidata/Wikipedia curated + linked where notable
E1
40–80-word definition, correct syntax, inside first ~150 tokens, prompt-type tagged
E2
Comparative data in clean native tables; survives Markdown render
1,400+ tracked tasks across a 27-sheet execution checklist.
The companion SEO+GEO 24-Week Execution Checklist decomposes every week of this plan into 50–90 granular, trackable tasks - 1,400+ in total - with dropdowns, conditional formatting and a live completion dashboard. We run the plan from the narrative document; we execute and track it in the workbook.
Structure: ① Read Me · ② Dashboard · ③ Risk Register · W01–W24 (one tab per week)
TrackSOP referencePriorityAssigned-ToStatusDueDefinition of DoneComments
Ready to run the protocol
See exactly what we'd run for your business.
No black boxes. Book a strategy session and we'll walk you through the exact 24-week plan, scoped to your category.
Why protocols matter
Whatthe24-weekprotocolisbuilttoprevent
The EliteLeadGrow protocol is a phase-gated, 24-week operating plan for SEO and GEO work. Six phases, each ending in a go or no-go gate, take a site from crawler access and locked baselines through schema, answer-first content, earned authority and re-audit to a documented handover. Every number is sourced or struck, and every risky change has a rollback trigger.
Whatisaphasegateandwhydoestheordermatter?
A phase gate is a condition that must be true before the next phase starts, and the order follows dependency. If a crawler cannot reach a rendered page with a 200 status, schema, content and links built on top of it cannot be read. So access and baselines come first, entity data second, content third and outside authority fourth.
Each gate also guards against one specific failure, which is the real reason it exists.
Phase 1, week 4
Phase 2, week 8
Phase 3, week 12
Phase 4, week 16
Phase 5, week 20
The six phase gates and what each protects against
Phase and week
Gate condition (short form)
What it protects against
Phase 1, week 4
Target AI and search agents return 200 and are logged; baselines locked
Optimizing a site crawlers cannot read
Phase 2, week 8
Schema validates clean; entity consistent across controlled surfaces
Contradictory facts about the business
Phase 3, week 12
Top pages re-engineered and extraction-tested; trust and author signals live
Publishing volume without quality
Phase 4, week 16
First earned citations landing; directories consistent; review flow compliant
Manufactured authority
Phase 5, week 20
Positive share-of-voice change against the week 4 baseline on the locked prompt set
Claiming progress nobody measured
Phase 6, week 24
Repeatable offering shipped; final report and dashboard handed over; quarterly cadence scheduled
Work only we can repeat
Whylockbaselinesbeforechanginganything?
Without a timestamped starting measurement, an improvement cannot be separated from drift, seasonality or luck. The protocol locks two baselines in weeks 1 to 4: the classic search footprint (landing pages, ranking keywords, referring domains) and share of voice on a fixed diagnostic prompt set.
The prompt set is a small, version-controlled list of the real questions buyers ask ChatGPT, Perplexity, Gemini and Claude, tagged by type. Share of voice is the percentage of those prompts where the brand appears. The set stays fixed because AI answers vary from run to run, and a swapped question makes two measurements incomparable. If share of voice swings sharply with no content change, the protocol's own rule is to audit the prompt set and rebaseline before reporting anything.
ChatGPT
Gemini
Claude
Perplexity
Your business
Howdoestheprotocolhandlerisk?
It decides rollbacks in advance. Every high-risk change has a trigger, an action and an owner written down before the change ships. The two riskiest activities are the week 3 change that opens crawler access through the firewall and the week 6 schema deployment.
A firewall rule that lets AI crawlers in can also block real visitors, and broken structured data can cost a site its rich results. Each change carries a monitored trigger, such as a conversion drop after a firewall change or a spike in Search Console coverage errors after a schema deploy, and a defined reversal: revert first, then tell the client.
Pre-agreed triggers matter because an incident is the worst time to debate whether to roll back. Documentation is versioned, and a peer or lead reviews the work before every client deploy.
Whatruleskeeptheworkhonest?
Three standing principles apply to every engagement, and each one protects you from paying for something that does not work or could harm you.
llms.txt is hygiene. It is shipped for agent and developer routing and is not sold as a citation driver, so you are not paying for a lever the evidence does not support.
Off-site authority is earned and disclosed. Fake reviews and planted mentions carry legal exposure, including under the US FTC rule on fake reviews, and platforms increasingly detect them.
Every number is sourced or struck. A statistic without a primary source is removed or reframed as a hypothesis tested against your own baseline.
Rankings
AI mentions
Leads
Calls
Howshouldyoureadthebi-weeklyscorecard?
Read deliverables against plan first, then share of voice, and treat referral traffic as only part of the picture. A one-page scorecard arrives at the end of every even week with the same six lines each time: share-of-voice change, source-inclusion change, AI referral sessions, deliverables shipped against plan, displacement wins and risks with the next fortnight's plan.
Deliverables against plan show whether the work happened. You can verify them directly.
Share-of-voice change against the week 4 baseline is the headline. Small fortnight-to-fortnight wobble is expected because AI answers vary.
Source inclusion separates a mention from a mention with an active link to your domain.
AI referral sessions count visits that followed an AI answer. Many readers see an answer and search the brand later, so brand search growth in Search Console captures part of the effect.
A gate is met by something checkable: server logs, validator output or a dashboard, not a status message.
Compare with the baseline
Check the result against the locked, dated starting measurement and the same prompt set.
Approve or ask again
Sign off the next phase when the condition is true, or ask what is missing and when it will be.
Questions
Protocolquestions
The structure is. The content differs, because each engagement leads with the client's highest-value cluster of commercial terms and rotates through the next ones phase by phase.
No. Results are measured against the baseline locked in the first weeks, and SEO is measured in months. The protocol guarantees process and transparency, not a position.
On this page. The week-by-week plan, gates, risk register and measurement model are published above, including owners, dependencies and the definition of done for each week.
Search Console, Bing Webmaster Tools, GA4, Screaming Frog, schema validators, Looker Studio dashboards and AI-visibility trackers, with a fixed prompt set as the measuring instrument.
The final report and dashboard are handed over, the offering is documented so it can be repeated, and a quarterly cadence is scheduled.
Because the evidence that answer engines rely on it is thin. The protocol ships it for agent and developer routing and does not present it as a way to win citations.
A named point of contact, delegated access to your analytics and CMS, and sign-off from whoever manages your firewall or CDN, since crawler access changes touch infrastructure you control.