Which Questions Reveal Whether an SEO Company Can Actually Deliver?

Begin with one concrete deliverable that matters to your engagement, such as a technical audit, service page, content update, migration plan, or reporting cycle. Ask the company to walk through the work from intake to completion.

Who gathers the inputs? Who decides priority? Who performs the work? Who checks accuracy? What does the client approve? What evidence closes the task? A useful answer identifies roles, dependencies, quality gates, expected outputs, and the system of record.

For content, request a redacted brief that shows search intent, audience, source requirements, subject matter input, editorial review, compliance review where needed, and final approval. For technical work, request a sample issue record showing reproduction steps, affected templates or URLs, business risk, recommendation, implementation owner, test method, and rollback consideration.

For reporting, ask how discrepancies are investigated and how decisions are documented. The agency does not need to reveal another client's private data or every internal instruction. It should be able to provide a representative artifact or create a process map for your proposed engagement.

Also ask how the workflow changes when your team is late with access, an expert disputes a claim, a developer rejects a recommendation, or an implementation fails validation. The output of this interview stage is a responsibility map and a list of evidence you will receive during delivery. This lets you compare companies on operational clarity rather than presentation quality.

Modern search work includes more than selecting keywords, but the interview should avoid vague claims about controlling a Knowledge Graph. Ask the company to define the entities that matter to your business: the organization, founders or practitioners, services, products, genuine locations, certifications, publications, and core topics.

Then ask where the approved facts will be stored and who maintains them. In 2018, many sales conversations still centered on keyword density and isolated metadata changes; today, a capable provider should also be able to discuss content relationships, authorship, source consistency, and visible business information.

Request an example of how the agency decides whether Organization, Person, Article, Service, LocalBusiness, or another schema type fits a page. The answer should begin with visible content and documented eligibility, not with an unsupported promise that markup will produce rankings, a panel, or an AI citation.

Ask how sameAs references are selected and how the team prevents links to profiles that do not represent the same entity. For authors, ask how biographies, credentials, bylines, reviewer roles, and external records are reconciled.

For locations, require useful location-specific information and genuine operations before approving a dedicated page. The output should be an entity register with approved names, relationships, sources, owners, and review triggers.

This gives content, technical, public relations, and reporting teams one maintained record while leaving room for legitimate name variations.

For regulated or technically complex topics, the agency should not infer professional facts from a keyword list. Ask which information must come from your team, which sources the writer may use, and who has final authority over substantive claims.

A practical workflow may combine intake documents, recorded interviews with consent, written expert answers, source packets, existing policies, and review comments. The company should be able to explain how it prepares questions, separates fact from opinion, records jurisdiction or product limitations, and turns specialist language into content a reader can understand.

Ask what happens when the subject matter expert and SEO lead disagree. The answer should protect accuracy and professional obligations while allowing the team to document search implications and propose alternatives.

Also ask how the agency verifies changing regulations, guidance, prices, services, or clinical and legal statements before an update. AI may be used for outlining, transcription, classification, or drafting, but the interview should focus on controls: approved inputs, prohibited uses, fact checking, source retention, privacy, authorship, and human approval.

Request a redacted brief or revision history that shows where expert evidence entered the work. The output should identify the content owner, expert contributor, editor, compliance reviewer where required, and publishing approver. Without those roles, the client may receive polished text that nobody is prepared to defend.

Google's Search Generative Experience, or SGE, was a historical experimental name. For current discussions, ask how the agency observes Google AI Overviews and other AI search products without treating them as one system.

The company should define a query set, products to test, locations or account conditions where relevant, dates, cited sources, exact mention classifications, and accuracy checks. Ask how it separates an observed source pattern from an official ranking factor.

A sound plan may improve answer clarity, authorship, crawl access, source support, and entity consistency because those changes are useful to readers and retrieval systems, but it should not promise a carousel position, citation, or recommendation.

Challenge claims about 'chunkability,' Speakable markup, FAQ markup, posting cadence, or special LLM feeds. The agency should show documented product guidance or label the idea as a test. Structured data must match visible content and supported types; it is not a guaranteed feed into a model knowledge base.

Ask how the agency records whether the brand was mentioned, cited, recommended, compared, or omitted, and whether the description was accurate. For AI-response studies, the report should preserve the exact recorded recommendation classification rather than inventing a purchase or hiring event.

The output is an AI visibility test register with queries, observations, sources, limitations, owners, and next decisions. That is more useful than relabeling ordinary SEO tasks with a new acronym.

A crawler report is an input, not a technical strategy. Ask the company to explain how it moves from detected issues to prioritized work. The decision criteria may include affected templates, indexation or rendering impact, conversion importance, traffic exposure, implementation effort, dependency, reversibility, and confidence in the diagnosis.

For a large or changing site, ask whether server logs, crawl samples, rendered output, analytics, Search Console, templates, and release history are needed to confirm the problem. Do not require log analysis when it adds no decision value, but confirm that the team knows when it is appropriate.

Ask how JavaScript content, canonicalization, redirects, pagination, internal links, faceted navigation, sitemaps, international targeting, and Core Web Vitals are evaluated in your specific stack. For a migration or redesign, request the sequence for inventory, mapping, staging tests, launch controls, rollback, monitoring, and post-launch validation.

Monitoring should name the event, threshold, owner, escalation path, and expected response. A generic promise of automated alerts is not enough. The technical roadmap should distinguish urgent risk reduction, foundational improvements, growth dependencies, and optional enhancements.

It should also identify who can implement each item and what happens when the recommendation conflicts with product, security, design, or engineering constraints. The output is a prioritized change plan that your team can approve, implement, test, and revisit.

Reporting should help you decide what to continue, change, stop, or investigate. Begin by defining the business actions that matter: qualified form submissions, calls, bookings, applications, purchases, demos, retained clients, or another outcome appropriate to the business.

Then ask which events can be measured accurately, which require customer relationship management or offline data, and which will remain estimates. The company should distinguish discovery metrics, engagement, conversion, lead quality, pipeline, and revenue.

Ask how GA4 events are specified, tested, documented, and changed. For calls or offline conversions, confirm consent, privacy, retention, access, and the method used to connect the event to a landing page or campaign.

Multi-channel attribution should be described as a model with limitations, not as perfect knowledge of causation. Request a sample report that explains what changed, why the team believes it changed, alternative explanations, data quality issues, and the next decision.

Cost per lead can be useful only when cost and lead definitions are complete and comparable. Ask how junk traffic, spam forms, internal visits, bots, duplicate leads, and irrelevant queries are filtered or flagged.

The reporting owner should reconcile data sources and document material discrepancies. The output is a measurement specification and reporting cadence tied to decisions, not a collection of charts designed to make every month look positive.

Most interview guides overvalue portfolios, rankings, and broad client lists. Those materials can establish context, but they rarely show who performed the work, which constraints applied, whether the outcome was typical, or whether the same process fits your site.

Another weak question is 'how many links will I get?' Link volume does not reveal relevance, editorial legitimacy, risk, destination choice, or how the agency will handle a questionable placement. The better request is for the decision criteria and controls behind off-site work.

Likewise, asking to see every internal document is unreasonable and can conflict with confidentiality or security. You need enough evidence to evaluate the operating method: a redacted example, process map, role definition, quality gate, or validation record.

A credible agency should be able to explain its work without disclosing another client's data. The interview should therefore test transparency, not demand unrestricted access; test repeatability, not a rigid template; and test business fit, not isolated historical results.

Earlier in my career, I placed too much weight on clever tactics and confident predictions. Those signals were easy to sell and difficult for a client to audit. The more durable distinction is operational: does the company define the problem, preserve evidence, assign ownership, document tradeoffs, validate implementation, and report limitations?

A good process is not automatically slow or bureaucratic. It should make routine work faster while reserving deeper review for high-risk decisions. I have also learned that transparency does not require an agency to surrender confidential methods or another client's information.

It requires enough explanation and representative evidence for the buyer to understand what will happen, who is responsible, and how quality will be checked. When interviewing a company, listen for precise answers about inputs, decisions, dependencies, and failure handling. A persuasive promise can be copied. An accountable operating process is harder to imitate and easier to evaluate.

Days 1-5

Document your current brand, key experts, services, genuine locations, critical pages, business goals, known risks, and access constraints before speaking with agencies.

Outcome: A shared baseline that lets candidates evaluate the same entity, scope, and business context.

Days 6-10

Draft 5-10 niche questions covering customer decisions, terminology, claim risks, regulations, and operational details that generic writers would struggle to answer accurately.

Outcome: A practical test of how each company gathers, verifies, and applies subject matter expertise.

Days 11-20

Interview 3 agencies using the same process, entity, expert, AI search, technical, and measurement questions. Request representative redacted evidence rather than confidential client files.

Outcome: A comparable scorecard showing process clarity, ownership, evidence quality, risk control, and fit.

Days 21-25

Ask the top 2 candidates for a preliminary technical and strategic roadmap that states assumptions, priorities, dependencies, exclusions, owners, and validation methods.

Outcome: Evidence of how each candidate turns limited discovery data into a qualified, business-aware plan.

Days 26-30

Check references for process reliability, communication, issue escalation, implementation quality, reporting honesty, and change handling rather than only ranking success.

Outcome: A final partner decision supported by consistent interviews, artifacts, references, risks, and operating fit.