TL;DR
- Define the problem and required outputs before comparing agencies. An audit, a content draft and a deployed change are different deliverables.
- Ask for evidence with clear samples, denominators, reporting conditions and attribution limits; inspect an authorized reporting example.
- Compare proposals on the same scope, including client inputs, approvals, ownership, reporting access and exit handoff.
- Question permanent ranking promises and unsupported results. If the offer, evidence or internal owner is unresolved, address that before buying a broad engagement.
Define the problem before asking for a proposal.
Write down the offer, markets, languages and decisions you need to support. State whether the issue is unknown visibility, incorrect brand facts, weak decision pages, technical access or poor inquiry quality. “Do GEO” is not a reviewable brief.
An audit identifies findings and priorities. A brief specifies content. A draft is written content. Implementation changes the live site. Confirm which of these the proposal includes, what is excluded and what your team must do.
Ask each provider the same questions below. These are editorial procurement prompts. A missing answer is a reason to clarify scope, not proof of misconduct.
| Question to ask | Credible answer / evidence | Red flag | Follow-up |
|---|---|---|---|
| Which markets, languages, offers and platforms are included? | Named boundaries and explicit exclusions | “All AI platforms” without a defined list | Write the agreed coverage into the scope |
| How is the prompt sample and baseline chosen? | Exact prompts, clusters, run conditions, denominator and saved outputs | Only favorable screenshots or a score without a sample | Ask to reproduce a small authorized sample |
| What does the technical review inspect? | Named pages/checks with observed responses and recommendations | Only “add schema/llms.txt” without a diagnosed problem | Request one finding traced to evidence and acceptance |
| Do we get advice, briefs, drafts or deployed changes? | Deliverable formats and publishing responsibilities | “Content optimization” hides the actual output | List files, revisions and publication ownership |
| What do the case metrics measure? | Period, baseline, unit, source and attribution limits | All website revenue counted as GEO contribution | Ask what other work or campaigns changed |
| What must our team provide? | Evidence, factual approval, access and development requirements | No client inputs despite specialist claims | Name the internal owner and dependencies |
| Can we access reporting evidence? | Raw or appropriately redacted records and export terms | Only a dashboard score with no counting rule | Walk through numerator, denominator and exclusions |
| Who owns content, data and accounts? | Clear ownership/access terms and approved permissions | Ownership or access is left unstated | Confirm assets and access at each milestone |
| How is third-party publication handled? | Authorized placements, disclosed commercial relationships and platform-policy checks | Fabricated reviews, bulk fake accounts or undisclosed planting | Ask for the method and approval process before authorizing it |
| What happens when we stop? | Handoff of approved work, source records, access and open tasks | No usable export or exit plan | Agree handoff format and responsibility in advance |
Inspect the evidence behind the pitch.
Read a case study as an account of a specific project. Check whether its headline concerns impressions, clicks, inquiries or revenue and whether the periods are comparable. An anonymous case can be useful if its measurement method is clear; a named logo alone cannot supply that method.
Request an authorized, redacted sample. Do not ask the provider to expose another client’s private information. A demonstration should be labeled as such, and project-reported results should not be presented as independent causal proof.
A useful report includes the question set, eligible-answer counts, failure/exclusion counts, dates, platforms and modes, source links, label rules and unresolved errors. It separates identifiable visits and qualified inquiries from sampled answer metrics. Use the measurement definitions to question any unclear rate.
Compare two proposals on equal terms.
Illustrative example with fictional proposals and no price claims: Proposal A supplies an audit and recommendations. Proposal B also includes approved briefs, drafts, implementation and follow-up checks. A lower fee for A may reflect less work, not better value.
First choose a required scope. Then ask both providers to quote that same scope or show the missing work and the cost of doing it internally.
| Required work | Proposal A | Proposal B | Buyer’s clarification |
|---|---|---|---|
| Research and baseline | Included | Included | Are platforms and sampling conditions comparable? |
| Content changes | Recommendations only | Briefs and drafts | Who validates facts and how many revisions are included? |
| Publishing | Client team | Included for agreed pages | What access, approval and developer support are needed? |
| Follow-up | Initial report only | Scheduled re-test | Does the repeat use the same sample and record failures? |
If your team already has capable writers and developers, A may be the right purchase. If it lacks delivery capacity, B may cover more of the real need. Neither label establishes quality. Judge the evidence and the total work required.
What changes scope and cost?
Markets, languages, offer complexity, sample breadth, repeat frequency, source review depth, content volume, implementation access and approval rounds all affect the work. Ask which assumptions drive the fee and what would trigger a scope change.
Get clear terms for content ownership, account access, reporting exports and exit support. Do not grant broad access simply because it is convenient; agree the access actually needed. This guide does not set a market price or alter SemShift’s published offer.
Distinguish warning signs from transparent commitments.
A provider can commit to deliverables, review dates and implementation checks. It cannot credibly promise a permanent first position across every independent AI platform. Ask how any outcome promise is defined and supported.
Missing denominators, unknown collection conditions and claims that all revenue came from GEO make results difficult to verify. Fabricated reviews or hidden promotional identities are a different concern: the acquisition method itself needs scrutiny. Paid content and anonymous cases are not automatically invalid; the relevant questions are disclosure, authorization, source quality and measurement.
Google cautions against guaranteed search rankings when selecting an SEO provider. Our analogous procurement question for GEO is what the supplier actually controls. Google’s people-first guidance also favors clear sourcing; it does not certify an agency or its claims.
When an agency is not the next step.
Resolve a disputed offer, absent evidence or missing internal approval owner before buying a large content program. If you already have a clear task and capable delivery team, targeted technical help may fit better than a broad retainer. If the bottleneck is lead handling, another visibility dashboard may not address it.
The implementation plan can clarify the work you want to buy. Before contacting a provider, prepare your website, offer, target markets, one documented problem, current reporting and available internal support. Then compare SemShift’s actual service scopes.