A small U.S. B2B service business should choose an AI-search agency as a measurement-led SEO and authority partner, not as a vendor promising guaranteed ChatGPT rankings. The best agency will improve the fundamentals that influence both conventional search and AI answers—useful content, technical accessibility, clear business entities, credible third-party references, and conversion tracking—while distinguishing evidence from speculation. What “AI visibility” should mean Ask agencies to define visibility in terms of buyer questions, not a single proprietary score. Your measurement plan should track: Whether your company is mentioned for relevant problem, category, comparison, and “best provider” prompts. Whether AI systems describe your services accurately. Which pages and third-party sources are cited. Visibility across the platforms your buyers actually use, such as Google AI features, ChatGPT Search, Perplexity, Gemini, or Microsoft Copilot. Qualified visits, calls, form fills, booked meetings, opportunities, and revenue influenced by organic and AI-assisted discovery. Be cautious of claims that a company can “rank ChatGPT.” Google says AI Overviews and AI Mode still depend on pages being indexed and eligible for ordinary Google Search, and that there is no special AI schema required. It also explicitly warns against over-focusing on structured data. OpenAI separately documents OAI-SearchBot as the crawler used to surface sites in ChatGPT search results, so crawler access and normal technical SEO are practical checks—but neither guarantees a citation. developers.google +2 Evaluation criteria Use a weighted scorecard rather than choosing from agency “top 10” lists. Criterion What to look for Evidence to request Relevant B2B experience Similar service, deal size, geography, sales cycle, and buyer roles Two references and a case study with starting point, work performed, timeframe, and outcomes Business understanding Research into customer problems, buying committees, competitors, and commercial terms A sample discovery plan and proposed buyer-question map Technical SEO Indexability, internal linking, page quality, structured data where appropriate, analytics, and crawl controls A redacted technical audit showing prioritized fixes—not a generic checklist Content quality Original expertise, service pages, comparison pages, FAQs, proof, and content reviewed by subject-matter experts One proposed outline and examples of published work Entity and authority building Consistent company information and credible mentions in relevant industry, local, association, and editorial sources Specific target sources and a clear outreach/PR process AI-search measurement Repeatable prompts, multiple engines, citation/source tracking, accuracy checks, and change-over-time reporting A sample report containing prompts, dates, raw outputs, citations, and methodology Revenue attribution Connection to CRM, call tracking, forms, assisted conversions, and pipeline A measurement specification showing how leads move from search to opportunity Safety and compliance No fabricated claims, undisclosed paid links, fake reviews, copied content, or confidential-data exposure Written quality-control, approval, privacy, and escalation procedures Commercial fit A senior person actually doing the work, manageable scope, ownership of assets, and reasonable exit terms Named team, deliverables, hours or capacity, contract, and account-access policy Treat public agency rankings as lead-generation material rather than independent proof. For example, recent industry roundups emphasize AI share of voice, citation authority, structured data, and pipeline attribution, but these are still claims made by third parties or agencies and should be verified through references and account evidence. finance.yahoo +1 Questions to ask finalists Ask each agency to answer these in writing: “Show us one comparable client before and after.” What changed in qualified pipeline, not just impressions or AI mentions? “How do you test AI visibility?” Ask for the exact prompt set, locations, platforms, sampling frequency, and treatment of personalized or variable answers. “How do you separate causation from correlation?” An increase in citations may reflect an algorithm change, brand activity, or normal SEO growth. “What would you not do?” A credible agency should reject guarantees, mass-produced pages, fake reviews, undisclosed link schemes, and fabricated expertise. “Which sources should validate us?” The answer should identify relevant independent publications, associations, directories, partners, experts, and customer evidence—not merely more pages on your own site. “Who owns the work?” Confirm that your business owns the website, content, analytics accounts, dashboards, prompt data, and outreach-created assets. “How will you use AI internally?” Require human review, fact-checking, source controls, confidentiality protections, and disclosure where appropriate. “What happens if results are flat after six months?” Look for agreed milestones, diagnostic reviews, and the ability to reduce or terminate scope. Red flags Reject or heavily discount an agency that: Guarantees placement in ChatGPT, Google AI Overviews, or any other answer engine. Reports only an unexplained “AI visibility score,” screenshots, rankings, or traffic. Cannot provide raw prompts, dates, citations, and the platforms tested. Equates more AI-generated pages with a strategy. Promises hundreds of backlinks, “secret” methods, or guaranteed featured snippets. Uses fake authors, invented statistics, unverified customer claims, or automated content published without expert review. Wants to control your domain, Google Analytics, Search Console, CRM, or advertising accounts. Avoids discussing negative or inaccurate AI descriptions of your business. Has no plan for measuring calls, booked meetings, opportunities, and revenue. Google’s spam policies specifically identify manipulative link practices and large-scale production of unoriginal content created mainly to manipulate rankings as violations. That makes “content volume” and cheap link packages particularly poor proxies for AI-search competence. developers.google A practical selection process Define 20–40 buyer prompts. Include local intent, urgent problems, service comparisons, vendor-selection questions, pricing/process questions, and questions involving competitors. Establish a baseline. Record current rankings, indexed pages, branded and non-branded traffic, conversions, CRM pipeline, mentions, citations, and AI answers. Save the prompts and outputs so later comparisons are reproducible. Invite three agencies. Give them the same brief and ask for a 90-day plan, assumptions, deliverables, risks, team, cost, and measurement design. Score them. A reasonable small-business weighting is: relevant proof 25%, measurement and attribution 20%, strategy and buyer understanding 20%, technical/content execution 15%, authority approach 10%, commercial and risk fit 10%. Run a paid diagnostic or pilot. Require a prioritized audit, prompt baseline, technical fixes, one or two high-value content assets, and a reporting prototype. Avoid a long lock-in before seeing how the team works. Review monthly, decide quarterly. Evaluate progress against leading indicators—technical fixes, useful assets, relevant citations, qualified visits—and lagging indicators such as opportunities and revenue. For many small businesses, the right arrangement is an existing strong B2B SEO/content agency with genuine AI-search measurement capability, rather than a specialist selling “GEO” as a separate magic layer. Google’s current guidance continues to emphasize valuable, distinctive content and foundational SEO for generative search features. The agency should also handle AI-related risks responsibly: NIST’s AI Risk Management Framework highlights reliability, security, accountability, transparency, privacy, and bias management as trustworthiness considerations. developers.google +2 Bottom line: select the agency that can show comparable B2B outcomes, explain its methodology in testable terms, preserve your ownership and reputation, and connect search visibility to qualified pipeline. A small pilot with transparent raw data is usually more informative than a polished AI-search promise or an industry ranking.