Can This SEO Contract Survive Due Diligence?

Start with ownership because it determines whether the campaign creates a durable company asset or a temporary vendor dependency. Ask: 'Which accounts, files, content, configurations, and third-party relationships will be created, and who will own each one?' Then ask: 'Which parts of the work stop functioning if the contract ends?' The agency should answer with an asset and access register rather than a broad assurance.

That register can cover the CMS, analytics, Search Console, tag management, local profiles, content files, design files, outreach records, reporting exports, custom scripts, structured data implementations, and any platform used to collect or publish material.

Ownership is not the same as access. Your company may own a website but still lack administrator credentials, recovery email control, billing access, or documentation needed to operate it. The contract should identify the client-side owner for each critical account and require agency access to be granted through named user permissions where the platform supports them.

There are legitimate tradeoffs. An agency may use licensed software, hosted reporting, or a managed technical layer to work faster. The issue is not whether a tool is proprietary. The issue is whether the dependency is disclosed, priced, exportable where relevant, and removable without damaging the client-controlled site.

For example, a reporting dashboard can disappear at termination if the underlying analytics accounts, definitions, and export files remain available. By contrast, core metadata or page content that vanishes with a vendor subscription creates a larger continuity risk.

Ask for the proposed account structure, the export format for campaign data, and the exact offboarding state. The decision output should be a contract schedule listing each material asset, its owner, its administrator, its storage location, its dependency, and its transfer or retention rule. This turns an abstract claim about 'building authority' into inspectable custody of the work actually produced.

The next contract risk is not content volume. It is unclear responsibility for factual accuracy, professional judgment, compliance review, and final publication. Ask: 'Who will research and draft each content type, and what evidence must they use?' Then ask: 'Who has authority to approve high-risk claims before publication?' A credible answer distinguishes roles instead of implying that one generalist writer can perform every function.

The agency may own keyword research, search intent analysis, outlines, editing, on-page implementation, and production management. Your internal specialists or designated external reviewers may need to validate statements that depend on legal interpretation, medical judgment, financial rules, product specifications, or current company policy.

The contract should identify which topics require that review and what happens when the reviewer disagrees with the draft. A useful sample workflow starts with a brief that records the target decision, intended reader, approved sources, claims requiring confirmation, internal expert, prohibited language, and conversion action.

Drafting follows the brief. Editorial review checks clarity, duplication, sourcing, and alignment with the page purpose. Subject matter review checks the claims within the reviewer's competence. The final approver accepts, rejects, or returns the page with comments.

Publication should not be treated as proof that every statement is permanently current, so the workflow also needs a method for correcting material errors and revisiting content when the underlying facts change.

Ask how AI-assisted drafting is governed. The relevant issue is not the label on the tool, but whether the agency can trace claims, detect unsupported additions, preserve confidential information, and obtain the required human approval.

Request a real sample brief, draft history, source record, and approval log with client details removed. The decision output should name the editorial owner, the reviewer for each risk class, the evidence expected, the approval state required for publication, and the escalation path for disputed or time-sensitive claims.

Monthly charts do not show whether the work was controlled. Ask: 'What record will show every material change made to our site or accounts?' Then ask: 'How will you connect that change to an expected outcome without presenting correlation as proof?' The agency should be able to show a sample work log that is detailed enough for another qualified person to understand what happened.

A useful entry records the affected URL or system, the issue or opportunity, the baseline evidence, the proposed change, the person responsible, the approver, the implementation date, the verification method, and any rollback instruction.

For content, it can include the previous version, revised version, source record, review status, and publication date. For technical work, it can include a ticket, code or configuration reference, test evidence, deployment status, and post-deployment check.

This documentation serves several owners. Marketing needs to know what was delivered. Technical staff need to know what entered production. Compliance or legal reviewers may need to understand the basis for a public claim.

Leadership needs a concise view of progress, risk, and next decisions. The contract should therefore distinguish the detailed operational log from the executive report. It should also define approval thresholds.

Routine metadata edits may follow an approved batch process, while migrations, large redirect changes, indexation controls, analytics changes, or sitewide templates may require explicit client approval.

Measurement should start with a recorded baseline and an expected mechanism, then compare relevant observations after implementation. Search performance can move for many reasons, so the agency should describe confidence and competing explanations instead of claiming that one task caused every change.

Request access to the project system or scheduled exports, but do not confuse real-time visibility with good documentation. A poorly structured activity feed is still difficult to audit. The output should be an agreed change record, decision log, issue register, and reporting cadence that your team can continue to use if account personnel change.

Technical delivery can be permanent, subscription-based, or a mixture of both. None of those models is automatically wrong, but the contract must make the dependency visible. Ask: 'Which implementations live in our CMS, codebase, accounts, or infrastructure, and which depend on your hosted service or license?' Then ask: 'What files, credentials, documentation, and assistance will be provided at termination?' A managed layer can be useful when it reduces deployment friction or provides specialist capability.

The tradeoff is continuity. If turning off that layer removes titles, redirects, internal links, structured data, analytics configuration, or page content, your team needs to know before signing. The contract should distinguish client-owned work product from agency tools and pre-existing intellectual property.

It should also state whether an export or migration path exists, who performs it, what format is delivered, and whether additional fees apply. Reversibility matters during the engagement as well. Large technical changes should have a rollback plan appropriate to the risk.

That does not mean every edit needs a complex release procedure, but sitewide templates, indexation directives, redirect sets, analytics changes, and code deployments should be testable and recoverable.

Before approval, ask the agency to walk through a hypothetical offboarding. The package may include an account and access inventory, current work log, open issue list, content and source files, configuration documentation, custom code covered by the agreement, reporting definitions, analytics annotations, placement records, renewal obligations, and a list of features that will stop.

Your internal technical owner should review whether the package is sufficient to keep the site operating. The commercial owner should check notice periods, transition assistance, data retention, deletion duties, and final billing.

The goal is not to threaten termination. It is to ensure that retention depends on ongoing value rather than an avoidable technical trap.

Google AI Overviews and other AI responses have changed what users may see before or alongside traditional results, but that does not create a guaranteed optimization channel. Ask: 'Which AI surfaces will you observe, and how will you record the result?' Then ask: 'What actions are supported by documented search guidance, and what actions are only tests or operating practices?' A credible agency should separate controllable work from external outcomes.

It can improve page clarity, factual support, crawlable access, internal organization, brand consistency, and the usefulness of content for the intended query. It cannot guarantee that a particular page or company will be cited, mentioned, summarized, or recommended in an AI response.

There is no special markup that guarantees inclusion in Google AI Overviews. Structured data can help search systems understand eligible page content when it accurately represents what users can see, but the agency should not sell an undocumented tag as an AI placement mechanism.

Measurement also needs discipline. AI outputs can vary by prompt wording, time, location, account state, and product surface. The agency should define a repeatable prompt set, record the date and context, preserve the response evidence permitted by the tool, and classify the observation precisely.

Useful classifications might distinguish cited source, unlinked mention, recorded recommendation, neutral description, competitor inclusion, or absence. A recorded recommendation classification is not evidence that a user hired the company.

Trend reporting should therefore show observations across a stable test set and explain method changes. Ask how the findings will change the work. For example, an inaccurate brand description may trigger entity consistency checks and source review.

Missing coverage for a recurring comparison question may trigger a useful page or section, provided it serves readers rather than merely duplicating query variants. The output should be an AI visibility measurement note with defined surfaces, prompts, evidence, classifications, limitations, owners, and actions, integrated with ordinary search and business reporting rather than presented as a guaranteed new ranking system. The shift is important, but claims about it should be more precise than they were a decade ago.

A reporting package can contain a large amount of data and still fail to answer whether the engagement is useful. Ask: 'Which user actions and business outcomes will define qualified progress for us?' Then ask: 'What can your attribution method support, and where will it remain uncertain?' Start by defining the decision the report must support.

Leadership may need to decide whether to continue, expand, redirect, or stop a workstream. Marketing may need to decide which topics and landing pages deserve further investment. Sales or intake teams may need to decide whether organic inquiries match the intended service, geography, and customer profile.

Those decisions require more than total sessions. A million low-intent visits can be less useful than a smaller set of visits that reach relevant service pages, complete meaningful actions, or produce qualified inquiries.

The agency should not invent revenue attribution it cannot observe. It should document analytics events, call or form tracking where appropriate, CRM handoff assumptions, consent and privacy constraints, duplicate handling, and the point at which responsibility moves to the client's sales or intake process.

For longer decisions, a search visit may assist an inquiry three months later, but that possibility does not prove a specific causal path. Reporting should label direct observations, modeled or assisted attribution, client-supplied outcomes, and unresolved gaps separately.

Geographic reporting should also reflect the business. A real office can justify a useful location page with location-specific staff, services, proof, directions, or operational details. A nominal market or broad service area does not automatically require a dedicated page.

Agree on exclusions for bot activity, irrelevant countries, job seekers, existing customers, or other segments only when the exclusion serves the stated decision and can be implemented responsibly. The output should be a measurement dictionary, baseline, dashboard, narrative interpretation, issue list, and next-decision section.

Each metric needs an owner, source, definition, review frequency, and action threshold. The agency should be judged on the quality of decisions enabled as well as the quantity of work completed.

Most buying guides still center the conversation on a client list, page one promises, or a list of tools. That creates two reasons for concern. First, references and case studies show that work happened elsewhere, but they do not define what will happen on your account.

Second, no agency controls Google's systems, so a ranking guarantee is not a reliable substitute for delivery terms. The stronger test is whether the agency can show a repeatable workflow and translate it into contractual obligations.

Ask how opportunities are selected, how changes are approved, how entity information is maintained, how subject matter expertise enters the process, how risk is escalated, and how outcomes are separated from activity.

Also inspect the technical debt the engagement could leave behind. A proposal can look productive while depending on agency-owned accounts, scripts, dashboards, or publishing processes that your team cannot operate.

Due diligence should end with a written decision record: accepted scope, rejected assumptions, unresolved risks, named owners, evidence required at each stage, and exit deliverables.

Early in my work, I gave too much weight to presentation quality, awards, and confident forecasts. Those signals can show that an agency knows how to sell, but they do not show how the account will be operated.

The more reliable evidence is quieter: a clear work log, disciplined research, honest uncertainty, named reviewers, controlled access, and a contract that does not depend on preventable lock-in. Building durable search visibility usually involves repeated technical checks, content refinement, internal coordination, and correction of weak assumptions.

It is not improved by hiding those steps behind a slogan. I now look for a partner that can explain the system before claiming the outcome. In high-trust work, restraint is part of performance. A single aggressive tactic, unsupported statement, or uncontrolled deployment can create work that outweighs the expected benefit.

The practical standard is therefore reviewability: can the client see what was decided, what evidence supported it, who approved it, what changed, what happened next, and what remains uncertain? That standard does not eliminate search volatility, but it gives the company a usable record and a way to improve decisions over time.

Days 1-5

Request the proposed tool, account, asset, access, and dependency register, including all proprietary services.

Outcome: A written view of ownership, administrators, vendor lock-in, export limits, and technical debt.

Days 6-12

Review a sample brief, source record, approval history, and the agency's process for integrating subject matter expertise.

Outcome: Evidence that content risk, factual review, and publishing authority match the complexity of your niche.

Days 13-20

Negotiate offboarding terms and inspect a sample technical change log, rollback note, and transfer package.

Outcome: Contractual and operational protection for client-controlled accounts, code, content, data, and documentation.

Days 21-30

Finalize the measurement dictionary, reporting examples, attribution boundaries, decision owners, and success criteria.

Outcome: A shared definition of useful progress that goes beyond vanity metrics and unsupported outcome claims.