Quick answer: a complete generative engine optimisation engagement covers five workstreams — baseline measurement, entity and structured data, answer-ready content, citation acquisition, and ongoing prompt-level tracking. Proposals that omit citation acquisition are the most common gap, and it is the workstream that does most of the work.
The five workstreams
1. Baseline measurement
Deliverable: a documented record of how each engine currently answers a defined set of buying questions in your category, including which competitors are named instead of you.
This is both the diagnosis and the yardstick. Without it every later claim of improvement is unverifiable. It should be produced before any work starts, and you should be given the raw prompts and outputs, not just a summary score.
2. Entity and structured data
Deliverable: Organization schema resolving to a single consistent entity, Service and FAQPage markup on commercial pages, and reconciliation of your brand facts across every place they appear.
The reconciliation half matters more than the markup half and gets less attention. If your website, company register, LinkedIn page and directory listings disagree about your founding year or trading name, a model hedges — and a hedging model names somebody else.
3. Answer-ready content
Deliverable: comparison, pricing, eligibility and FAQ content structured so individual passages can be quoted without surrounding context.
The format constraint is real and specific: answer first, then context; self-contained sections; concrete figures rather than adjectives. A page can be excellent by editorial standards and useless for extraction.
4. Citation acquisition
Deliverable: earned mentions in the specific third-party sources your category's answers draw on.
This is the workstream that decides whether you are recommended rather than merely described accurately, and it is the one most often quietly absent from a scope of work. It is slow, it cannot be automated, and it is why the credible retainer bands are what they are. Ask for named target sources, not an outreach volume.
5. Prompt-level tracking
Deliverable: monthly re-testing of the agreed prompt set across each engine, reported as share of voice against named competitors.
Insist the prompt set is fixed and agreed in advance. A prompt set chosen after results are known is marketing, not measurement.
What moves the needle, ranked
| Workstream | Time to visible effect | Effect on being recommended |
|---|---|---|
| Entity reconciliation | Weeks | Foundational — nothing else works without it |
| Structured data | Weeks | Moderate; removes ambiguity rather than adding authority |
| Answer-ready content | 4–8 weeks | Moderate; makes you quotable once you are trusted |
| Citation acquisition | 3–6 months | High — this is the differentiator |
| Tracking | Immediate | None directly; it is how you steer |
The ordering explains a common disappointment. Engagements weighted towards content and schema produce fast, visible-looking progress and then plateau, because the brand became accurately described without becoming preferred.
Reading a statement of work
Four questions to put to any GEO proposal:
- Is citation acquisition named as a deliverable with target sources? If it appears only as "authority building", assume it is not happening.
- Is the baseline delivered to you as raw prompts and outputs? Summary scores hide the method.
- Are the schema changes named by type? "Structured data optimisation" is not a deliverable; "Organization, Service and FAQPage markup across 40 pages" is.
- Does the timeline distinguish fast from slow? A proposal promising recommendation gains in month one has not understood the mechanism.
What is usually out of scope (and should be priced separately)
- Native-language content for non-English markets — needs native writers, priced per market.
- Site migrations or replatforming, even where the current platform is the obstacle.
- Digital PR campaigns beyond category-relevant citation outreach.
- Paid search, which is a different channel with different economics.
Frequently asked questions
What is generative engine optimization?
The practice of optimising your brand, content and structured data so generative AI engines surface, cite and recommend you. Our complete GEO guide covers the mechanics.
How is it different from what an AI SEO agency sells?
"AI SEO" is the commercial label; GEO is the discipline. The distinction that matters is what is measured — see what an AI SEO agency actually does.
Can we do this without content production?
Partially. Entity, schema and citation work can proceed on your existing content, and for a site with good material already, that is often the efficient sequence.
How many prompts should be tracked?
Enough to be representative, few enough to be re-run honestly every month. In practice fifteen to forty for a single market, weighted towards buying-stage questions rather than definitional ones.
Related reading: what agencies charge and schema markup for GEO. Our AI SEO and GEO services are scoped exactly along the five workstreams above — with market-specific delivery for the US and UK.