What Questions Should You Ask an SEO Agency?

The 12 questions every prospective SEO agency must answer

These questions expose the difference between a tailored operating plan and a generic sales package. Ask each finalist the same questions, record the answers and request supporting examples.

1. What business outcome will SEO improve, and how will you measure it?
2. What do you need to audit before recommending a strategy?
3. Which problems would you investigate first on our website, and why?
4. What will you deliver during the first 30, 60 and 90 days?
5. How do you prioritize technical fixes, content and authority building?
6. Who will work on our account, and what work is outsourced?
7. How will you collaborate with our developers, writers and subject matter experts?
8. How do you acquire links and evaluate their quality and risk?
9. How will you report rankings, traffic, conversions and revenue?
10. How are you adapting SEO for AI Overviews, AI Mode, Copilot and ChatGPT?
11. What happens when performance declines or an experiment fails?
12. What do we own if the engagement ends?

A credible agency can answer in operational terms. It should identify dependencies, distinguish recommendations from implementation and explain where uncertainty remains. Google itself warns that no one can guarantee a number one ranking and that changes can take from hours to several months to show an effect.

Use this answer quality matrix

Do not score an agency on confidence alone. Score the specificity, verifiability and commercial relevance of its answers.

Questions about diagnosis, strategy and prioritization

Ask, What evidence will you collect before deciding what to change? A useful discovery process normally examines Google Search Console and analytics data, crawl paths, index coverage, rendering, templates, canonical tags, internal links, backlinks, competitors, content quality and conversion journeys. For an established site, it should also inspect historical migrations, traffic declines and previous agency work.

Then ask the agency to describe its prioritization model. A defensible model weighs likely impact, confidence, implementation effort, business importance, reversibility and dependencies. A blocked checkout template may deserve attention before 20 new articles. A site with thousands of weak, overlapping URLs may need consolidation and indexation control before further publishing.

Ask how the agency handles topical architecture. Strong answers may include hub-and-spoke relationships, query fanout, entity coverage, internal link paths and the distinction between creating, updating, combining and removing pages. It should also discuss content decay, strategic refresh cycles and controlled title or intent tests rather than assuming every performance problem requires a new URL.

Finally, request one redacted audit and one prioritized roadmap. Check whether each recommendation identifies evidence, affected pages, owner, expected outcome and validation method. A list of tool warnings is not a strategy.

Questions about technical SEO and implementation

Ask, Will you implement technical changes or only recommend them? Clarify who writes tickets, edits templates, tests releases and confirms that search engines can render and index the result. If your platform is unusual, ask for directly relevant experience rather than general technical claims.

The agency should be able to explain crawl prioritization, robots controls, XML sitemaps, canonical discipline, redirects, duplicate URLs, JavaScript rendering, faceted navigation and structured data. On a large site, ask whether log-file analysis would reveal wasted crawling, neglected sections or bot responses that a conventional crawler cannot show.

For performance work, ask which field metrics will define success. Current Core Web Vitals thresholds at the 75th percentile are LCP within 2.5 seconds, INP within 200 milliseconds and CLS within 0.1. Passing these thresholds is useful, but it does not guarantee rankings or conversions.

Ask how recommendations are validated after release. A sound sequence is to test in staging, preserve crawl and canonical signals, deploy under change control, request or await recrawling, inspect rendered pages, monitor Search Console and compare affected templates. Structured data must describe visible, accurate content. Markup should not be used to imply reviews, products or business facts that users cannot see.

Questions about content, authority and links

Ask, Why would your proposed page deserve to rank or be cited? The answer should go beyond keywords. Look for plans to add expert explanations, proprietary examples, original data, decision tools, comparison assets, statistics pages or other information that satisfies the query better than existing results.

Ask how writers obtain subject matter expertise and how factual claims are reviewed. Determine whether the agency will interview your experts, identify source provenance and update volatile facts. Also ask how it prevents cannibalization and scaled low-value publishing. Google’s spam policies apply to mass-produced pages created primarily to manipulate rankings, whether automation, people or both produced them.

For authority building, ask for the exact methods, approval rules and sample reports. Sustainable options include digital PR, link-intersect analysis, reclaiming unlinked brand mentions, expert contribution programs and promoting genuinely useful research or tools. Ask whether the agency pays publishers, controls linking sites or guarantees placement metrics.

Bulk paid links, private networks and placements selected only by a third-party authority score create substantial risk. Do not accept hacked links, hidden links, doorway pages, fabricated reviews, cloaking or deceptive redirects. If an agency proposes a gray-area tactic, require a written explanation of the potential gain, policy risk, reversibility and affected domain.

Questions about AI search, AEO and GEO

Ask the agency to define what it means by AI search optimization. Google states that the same foundational SEO practices apply to its AI features, with no special markup or AI file required. Pages still need to be crawlable, indexable, useful and eligible to appear in Search.

A capable agency should map important user journeys into likely query rewrites and follow-up questions, then make answers explicit and independently understandable. It should strengthen entity relationships, factual passages, comparisons, definitions, supporting evidence and source clarity. This improves retrieval and answer absorption without sacrificing the human reader.

Measurement remains imperfect. Ask for a stable, business-relevant prompt or query set; the markets, devices and systems tested; mention and citation tracking; linked referral sessions; and conventional search performance for the same topics. Reports should separate observed citations from tool estimates. They should not present one personalized answer snapshot as universal visibility.

Independent findings point in useful but incomplete directions. Ahrefs found substantial overlap between AI Overview citations and strong organic rankings in an analysis of 1.9 million citations. A 2026 study of 112 Product Hunt startups found referring domains predicted Perplexity visibility, while claimed GEO tactics did not show a significant correlation. These findings support strong SEO and authority foundations, but they do not prove a universal formula for AI citations.

Questions about KPIs, reporting and troubleshooting

Ask for a sample dashboard and a sample written analysis. The dashboard should distinguish business outcomes from diagnostic indicators.

• Business outcomes: qualified leads, ecommerce revenue, pipeline value, booked appointments, assisted conversions or profit where available.
• Search outcomes: nonbrand clicks, relevant landing page visibility, local actions, product discovery and conversions by query group.
• Operational measures: recommendations shipped, pages improved, indexation changes, tests completed and outreach results.
• AI observations: citations, mentions, linked referrals and visibility across a controlled query set, with methodological caveats.

Ask how attribution limitations will be handled. Search Console provides query and search performance data, while analytics describes behavior and conversions. Their totals can differ because the systems measure and process data differently. The agency should reconcile trends without pretending the tools are identical.

Use a diagnostic question: If organic traffic falls 30 percent next month, what will you do in the first 48 hours? A strong answer checks tracking and annotations, segments by page, query, country, device and search type, reviews technical releases, examines indexation and security, considers seasonality and checks confirmed or suspected search changes. Google recommends this type of segmented analysis with Search Console and Google Trends. An automatic conclusion that an algorithm update caused every decline is a warning sign.

Questions about people, process, pricing and ownership

Ask who sold the engagement, who leads it and who performs technical, editorial and outreach work. Request each person’s relevant experience and expected allocation. If contractors or offshore teams are involved, focus on disclosure, supervision, security and quality control rather than location.

Clarify the operating rhythm: meeting frequency, response times, ticket systems, approvals and escalation. Ask what your team must supply, such as developer capacity, expert interviews, legal review or analytics access. An agency cannot be accountable for implementation while remaining unable to publish, brief developers or obtain decisions.

For pricing, ask what is included, what triggers extra fees and how unused hours or delayed approvals are handled. Retainers suit continuous programs, projects suit bounded audits or migrations, and performance-linked arrangements require especially precise definitions and safeguards. Compare deliverables, expertise and implementation responsibility rather than choosing by monthly price alone.

Confirm in the contract that your company owns its domain, website, Search Console, analytics properties, content, creative assets, outreach records and credentials. Require a documented exit process, data export and removal of agency access. Ask whether the agency works with direct competitors and how it protects confidential information.

Red flags and a practical agency scorecard

Disqualify or investigate any agency that guarantees rankings, refuses to describe link methods, promises a fixed number of high-authority links, creates accounts it will not transfer, relies on fabricated case studies or insists that every service is proprietary and therefore unverifiable. Also question reports that celebrate impressions while qualified conversions deteriorate.

Score finalists from 0 to 5 in each category, then multiply by the weight. Require evidence for any score above 3.

After scoring, run reference calls with questions about missed deadlines, unsuccessful recommendations and conflict resolution. References selected only for exceptional wins will reveal less than a candid explanation of how the agency handled a difficult quarter.

What is proven, accepted in practice and still uncertain

Supported by official guidance and measurable standards

Search engines must be able to access, render, understand and index eligible pages. Useful original content, descriptive titles, sensible links, accurate structured data and accessible site architecture support discovery. Search Console and analytics can measure search exposure and onsite outcomes, although their datasets are not interchangeable. Manipulative links, cloaking, doorway pages and scaled low-value content can violate Google’s spam policies.

Broad practitioner consensus

Experienced practitioners generally prioritize technical access, intent satisfaction, topical relevance, editorial quality, genuine authority and accountable reporting over bulk content or link packages. They also track AI mentions and citations separately from rankings. Community discussions support these observations, but anecdotes do not establish causality.

Still uncertain or context dependent

No stable formula guarantees inclusion in AI-generated answers. Citation monitoring varies by tool, location, model and prompt wording. The incremental value of tactics labeled GEO remains contested, and correlations between links, rankings and AI citations do not prove direct causation. A trustworthy agency will test these areas while distinguishing observations, hypotheses and verified outcomes.