From Signed GEO Proposal to Delivery: The Agency Handoff

The gap between "client signed" and "delivery started" is where AI visibility retainers quietly go wrong. Here's the internal reconciliation that keeps the promise intact.

Updated on: 2026-10-29

A deal closes. The sales lead forwards the signed SOW to delivery with a line like "all yours, they're excited." Three weeks later the client asks why their brand still isn't being recommended by ChatGPT, and the delivery team learns for the first time that someone implied a guaranteed recommendation during the close call. Nobody wrote it down. Nobody reconciled it. Now it's a trust problem instead of a scope conversation.

That gap is the subject here. The audit tells you what to propose. The proposal becomes a signed scope. Onboarding gathers client inputs. But between signature and the first real piece of delivery work, there's a distinct internal job: making sure the signed scope, the things said during the sale, the delivery plan, the measurement assumptions, and your actual team capacity all agree before you tell the client work is underway. This is the handoff, and it belongs to the agency alone.

What follows is a practical synthesis drawn from project-management guidance and public agency handoff practice. It isn't an independently validated industry standard, and no evidence here confirms it as any single vendor's documented operating procedure. Treat it as a working process you adapt to the size of the engagement.

Why GEO handoffs break in ways ordinary projects don't

Most handoff failures are generic: lost verbal promises, fuzzy scope boundaries, dates agreed before anyone checked capacity. Those apply to GEO work too. But Generative Engine Optimization adds a specific hazard that a web-design or standard SEO handoff doesn't carry as sharply.

The word "visibility" hides at least four different things, and they are not interchangeable:

A brand mention in an AI answer
An active recommendation ("you should try X")
A linked citation to the client's own site
Provider coverage, meaning presence across ChatGPT, Claude, Perplexity, Gemini, and others

A brand can be recommended without its site being cited. It can be cited without being recommended. The distinction between these signals is easy to blur in a proposal and easy to misremember after a close call. If sales sold "visibility" and the client heard "recommendation" while the proposal measures "mentions," you have three different success definitions riding on one signature.

The second hazard is measurement instability. Generative answers vary across runs and over time. A preprint on citation variability cautions that a single-run estimate can look more precise than it is. So the pre-sale audit score is a directional snapshot, not a stable baseline you can inherit without rerunning it under agreed conditions. Treat it as emerging evidence rather than a settled sampling rule, but let it make you cautious about handing one audit number to delivery as gospel.

The original GEO research paper reported visibility gains of up to 40% in its evaluated setting, with results varying by domain. That is a research finding, not a forecast for your client. If it crept into a sales conversation as an expected outcome, flag it now, before delivery is quietly holding a number nobody can promise.

The handoff is not the kickoff

Keep three events separate.

Onboarding intake collects what you need from the client: access, logins, brand context, who approves what. The internal handoff is agency-only reconciliation, no client in the room. The client kickoff confirms the agreed plan and surfaces anything still open.

The handoff sits in the middle and does the uncomfortable work: checking that what was sold matches what can be delivered. If you merge it into kickoff, you end up renegotiating scope in front of the client, which is the worst possible place to discover that sales promised content production the retainer wasn't priced for.

The reconciliation workflow

Scale this to the engagement. A small monitoring retainer doesn't need a heavyweight charter; PMI's small-project guidance is explicit that scope, deliverables, roles, assumptions, and acceptance criteria can be sized to the project rather than forced into a formal document every time.

Step	What the agency does	Done when
Reconcile the signed record	Put the signed SOW and any later written approvals in the project record. Compare against sales notes, the audit, the presentation, and any recalled verbal commitments. Flag conflicts rather than treating the proposal as the only version of the truth.	Delivery can point to the controlling agreement and name every promise that still needs clarifying.
Translate the promise into deliverables	Convert "AI visibility improvement" into specific work: monitoring, prompt research, content drafting, publishing, technical recommendations, outreach, reporting. Specify quantities, cadence, formats, review rounds, and explicit exclusions.	Each deliverable has an owner, an expected output, and an agreed boundary.
Check assumptions and dependencies	Record what the timeline depends on: client access, CMS and analytics availability, approvals, implementation support, third-party response. Mark anything unconfirmed as an open item with an owner and a date.	No critical dependency is being treated as satisfied when it hasn't been confirmed.
Check people, capacity, price, timing	Name the account owner and delivery lead. Check promised dates against real availability and required skills. Make the commercial context visible to the right internal roles, especially if the estimate assumed a fixed fee or capped hours.	Delivery has a named accountable lead and a feasible schedule, or the date is clearly marked provisional.
Compare sold measure with observable evidence	Preserve the baseline conditions and define what you'll compare: prompt set, provider, market and language, date, run conditions, metrics. Separate mentions, recommendations, citations, traffic, and business outcomes.	The team knows exactly what it will report and what it cannot responsibly promise.
Hold the internal handoff	Get sales or account leadership and delivery looking at the same record. Discuss discrepancies and open items out loud. A short meeting works; no source sets a correct length.	Every unresolved issue has a decision, an owner, and a due date. Serious conflicts get escalated before kickoff.
Build the executable record	Set up milestones, tasks, owners, reporting cadence, client-side responsibilities, and the first deliverable. Keep one shared record that links back to the authoritative agreement instead of letting conflicting copies circulate.	The team can start the agreed work without reconstructing the sale from memory.
Use kickoff to confirm, not renegotiate	Introduce the delivery team, restate objectives and boundaries, confirm client responsibilities and communication, surface open items. If the client wants more, record its scope and assess fee and schedule impact before treating it as included.	Client and agency share one understanding of what starts now and what stays conditional.

The through-line is simple: a written baseline is the source of truth, and any later change gets documented, assessed for impact, and approved before it becomes authorized work. Scope creep usually starts as a helpful "sure, we can add that" with no impact check behind it.

The discrepancies worth hunting for

Reconciliation isn't copying CRM fields into a project tool. It's actively testing where the sale and the delivery plan might disagree. These are the mismatches most worth probing in GEO work specifically. They're plausible checks, not ranked by prevalence.

What to test	The mismatch it catches	How to resolve it
Proposal vs. informal promise	SOW lists monitoring; a sales call implied content production was included.	Capture the statement. Decide whether it was in scope, excluded, or needs a paid change before work proceeds.
"Visibility" means different things	Client expects a recommendation; proposal measures mentions or citations.	Name each metric and define its intended result separately. These are not synonyms.
Audit snapshot treated as baseline	A pitch audit becomes a guaranteed stable starting measurement.	Preserve its conditions and rerun under agreed delivery conditions. Answers vary by run and time.
Work product vs. outcome promise	Team promises content and reporting; client remembers a guaranteed recommendation or lead lift.	Commit to controllable work and evidence-based reporting, not provider-controlled results.
Research vs. implementation	"Fix technical issues" sits in sales notes, but signed scope only covers recommendations.	Clarify who implements and how much implementation effort is included. Don't silently convert guidance into execution.
Drafting vs. publishing	Sales read "content delivery" as a draft; client read it as a published page.	Specify format, asset count, editorial approvals, CMS publishing responsibility, and included revisions.
Monitoring scope drift	Prompt count, providers, markets, languages, or cadence in the proposal differ from the delivery setup.	Reconcile each parameter before the baseline. Record exclusions and any expansion rules.
Timeline vs. dependency	The close presentation named a launch date; access and reviews are unconfirmed.	Mark the date conditional or re-agree it after checking dependencies.
Underspecified success	"Improve AI presence" has no defined comparison, window, or acceptance condition.	State what the report will show and what counts as completed work.
Stakeholder mismatch	The person who signed isn't the person who approves content or reporting.	Name the decision maker, day-to-day contact, approver, and escalation path.
Fee vs. effort	Fixed fee priced for monitoring; delivery assumes content, outreach, and implementation too.	Review scope against commercial assumptions internally and renegotiate before the effort is absorbed unpriced.
Scope-change ambiguity	New requests arrive; nobody knows who can authorize added work or its impact.	Document who can request and approve changes and how fee, schedule, and deliverables get assessed first.

My read is that the "visibility means different things" row causes more slow-burn damage than the obvious fee mismatch. A fee problem surfaces in week one when someone looks at the hours. A definition problem hides until the first report, when the client reads "brand mention share up 12%" and asks why they still don't get recommended. By then you're defending work instead of reporting it.

A labeled example record

This is a fictional illustration for a small GEO engagement, not a real client, proposal, or guarantee of any tool's behavior. It shows the level of precision worth reaching before you call delivery ready.

Record field	Illustrative entry
Engagement objective	Establish a comparable baseline for the client's presence in answers to a defined set of non-branded buyer questions; use findings to prioritize agreed work.
Authoritative scope	Signed scope document is the source of truth, linked in the project record. Later change approvals recorded separately.
Included work	Run and review the agreed prompt set across named providers; prepare a baseline report; recommend a prioritized action list.
Excluded unless separately approved	Guaranteed citations or recommendations; unlimited prompt expansion; content publishing; third-party placements; technical implementation by the agency; revenue attribution without suitable client data.
Measurement conditions	Record prompt wording, provider, market and language, date, and actual answer evidence. Report mentions, recommendations, and citations separately.
Starting evidence	Attach the pre-sale audit, noting it is directional and must be rerun under agreed conditions before serving as the delivery baseline.
Assumptions and dependencies	Client supplies business context and access; reviewer returns approvals within the agreed window. Confirm, don't assume.
Owners	Account lead owns the commercial record; delivery lead owns the readiness check; named client contact owns inputs and approvals.
Timeline	Dates contingent on receiving access and approvals; any firm client-facing deadline must match signed scope and confirmed capacity.
Open items	List each unresolved question, its owner, and due date. Distinguish blockers from items resolvable at kickoff.
Change handling	Added providers, prompts, content production, implementation, or outreach require a documented scope and impact decision before work is added.

The charter-like fields follow PMI's small-project scaling; the measurement fields reflect the split between observable visibility signals and broader business outcomes.

Keep outputs, observations, and outcomes separate

This is where delivery discipline earns its keep over the length of a retainer. Three different things get confused constantly:

A completed article or a finished monitoring run is a delivery output. You control it. You can promise it.
A change in what AI answers say is a visibility observation. You influence it. You can't guarantee it, because independent providers control retrieval and generation.
Leads and revenue are business outcomes. They need separate analytics or CRM evidence, and you report them only when the data actually supports the connection.

The public agency AI visibility retainer playbook makes the same division and is blunt about not guaranteeing mentions, citations, rankings, traffic, or revenue. That's not caution for its own sake. It's the only honest position when a third party owns the answer engine. If your handoff record commits delivery to a provider-controlled result, you've baked a broken promise into the plan.

For the analytics side of the handoff, one useful note: Google says appearances in its AI features are counted in the Search Console Performance report's Web search type. That's helpful context when you're reconciling what the client's existing analytics can and can't show, but it doesn't mean every AI visibility metric is measured there.

Where tooling fits, and where it doesn't

A platform helps the mechanical parts of this: recurring monitoring across providers, prompt research, draft preparation, and report assembly. Glim is built around that recurring-delivery model, with agent workflows for monitoring, content, Reddit, and outreach, plus white-label reporting and a free prospecting tier that produces the pitch audit in the first place.

But the handoff is a judgment job, not a tooling job. No platform decides whether a verbal promise was in scope, reconciles a fee against assumed effort, or names who approves content. Those are agency decisions. The tool preserves baseline conditions and generates the reporting; the account lead and delivery lead own validation, prioritization, interpretation, and what the agency is willing to commit to. Treat the software as support for the body of work, not a substitute for the reconciliation.

FAQ

Isn't a formal handoff overkill for a small monitoring retainer?

No. Scale it down, don't skip it. For a lean engagement the whole thing can be a one-page record and a fifteen-minute conversation between sales and delivery. The point is that someone checks the sold promise against the delivery plan before work starts. The format should match the size; the check shouldn't disappear.

What if sales made a verbal promise that isn't in the signed scope?

Capture it in the record and decide its status before delivery proceeds. It was either included, explicitly excluded, or it needs a documented change. The dangerous move is leaving it unresolved so delivery discovers it when the client raises it. A promise nobody wrote down is still a promise in the client's memory.

Can we just reuse the pre-sale audit as the delivery baseline?

Treat it as directional, then validate it under your agreed conditions. Generative answers vary across runs and over time, so an audit run during the pitch may not reproduce exactly. Preserve its prompt wording, provider, market, and date, then rerun under the conditions you've committed to report against. That rerun becomes the baseline you measure from.

How do we handle a client who wants more work after the SOW is signed?

Record the new request as its own scope and assess fee and schedule impact before treating it as included. Decide in advance who on your side can authorize added work. Silently absorbing requests is how a priced retainer turns unprofitable without anyone noticing until the margin review.

Who should own the handoff record?

Split it. The account lead owns the commercial record and the controlling agreement. The delivery lead owns the internal readiness check. The named client contact owns inputs and approvals. One shared record, three clear owners, linked back to the signed scope so no conflicting copies circulate.

What I'd do first

Pull your last three signed GEO engagements and check one thing: does each one have a written, agreed definition of what "visibility" meant in that deal, separated into mention, recommendation, citation, and coverage? If any of them just say "improve AI visibility" with no defined comparison or metric, that's the gap to close before your next close call, not after it.