Direct answer: choose a GEO agency by its ability to demonstrate six connected proofs: map real commercial questions, establish technical eligibility, create useful original sources, strengthen verifiable authority, measure appearances reproducibly and convert discovery into qualified demand. No agency controls whether a model cites a page. Treat ranking or citation guarantees as a warning. Require a hypothesis, baseline, method, evidence and commercial consequence for every workstream.
GEO, AEO or generative engine optimization describes work that makes a company understandable, retrievable and useful in AI search experiences. The label does not create a channel independent from the web. Google says its generative experiences continue to rely on core Search systems and recommends sound SEO, original content and authentic authority. OpenAI gives publishers crawler guidance and referral parameters for ChatGPT. A provider should work with those documented conditions rather than myths.
The MAKINAI GEO Evidence-6
Compare agencies through six proofs: Demand, Eligibility, Source, Authority, Measurement and Conversion. Every proof needs an output, metric, owner and stated limitation. Then use a neutral test protocol: select prompts before work begins, run web-enabled checks, record date, model, response and cited URL and compare movement over time. The agency must accept negative results without changing the success criterion after the fact.
1. Demand Proof: which questions justify investment?
Strategy should begin with customer decisions rather than a generic keyword list. Request a map connecting question, persona, stage, market, language, service, required evidence and next action. Queries such as “which company should we hire,” “how should we compare proposals” and “what does it cost” have different value from broad definitions. The agency should prioritize coverage by commercial potential, fit and ability to publish a defensible answer.
- Evidence: prompt inventory in English, Portuguese and Spanish; intent classification; corresponding page or gap; current presence; sources dominating the answer; difficulty; likely value; priority rule. Red flag: hundreds of nearly identical prompts are counted as separate opportunities.
2. Eligibility Proof: can the site be discovered and understood?
Before producing content, the agency should audit crawling, indexing, rendering, HTTP status, canonical, hreflang, sitemaps, internal structure, performance, mobile experience and appropriate structured data. For ChatGPT, OpenAI’s documentation says OAI-SearchBot should not be blocked when a publisher wants content considered for summaries and snippets. Crawler access establishes eligibility; it does not guarantee citation.
Require public validation of critical samples and a blocker list with impact and owner. Special files or invented markup do not replace accessible HTML and indexable content. Google explicitly says llms.txt and special AI markup are unnecessary for its generative search experiences. A credible partner distinguishes a documented requirement, an experiment and a hypothesis.
3. Source Proof: is there something worth citing?
Models and search systems need pages that answer clearly and contribute value. The agency should produce research, comparisons, frameworks, original data, documentation, tools, explanations and service pages grounded in real expertise. Content that merely rearranges consensus provides little selection reason. Require a direct answer, appropriate authorship, verifiable sources, freshness and a traceable relationship between claims and evidence.
Google recommends non-commodity, helpful and expert-led content. It also warns that producing many pages without user value can constitute scaled content abuse regardless of how pages were created. Ask for the original contribution behind every brief and which topics the agency rejected for lack of evidence. Volume should not be the primary KPI.
4. Authority Proof: do external sources confirm the entity?
A company website declares what the business is; independent sources help systems and people verify it. The proposal should map factual presence in relevant directories, associations, media, profiles, partners, events and third-party content. Require consistency in name, URL, description, leadership, services and markets, plus structured data connecting the entity to the same references.
Reject purchased mentions, artificial reviews, disguised sponsored lists and placement on unrelated domains. Google says pursuing inauthentic mentions is not a useful generative-search strategy. The agency should create legitimate opportunities through original research, editorial collaboration, genuine reviews, expert contribution and assets others genuinely want to reference.
5. Measurement Proof: can the result be reproduced?
Measurement should separate eligibility, indexing, impression, appearance, citation, click, session and lead. For prompt tests, define exact question, language, market, tool, model, web-enabled state, date and success rule. Count an appearance only when the answer uses or cites a confirmed company URL. Preserve evidence and history; one screenshot does not demonstrate stable coverage.
Google’s dedicated Search Console generative AI report shows impressions in AI Overviews and AI Mode when available. OpenAI says ChatGPT referral links include utm_source=chatgpt.com, enabling analytics measurement. The agency must state gaps: some platforms provide no impression or prompt detail, personalization varies and answers change. Do not accept a dashboard that turns estimates into facts.
6. Conversion Proof: does discovery become an opportunity?
A citation without a next action may have little business value. Verify that citable pages connect readers to an appropriate service, diagnostic, contact path or tool. The agency should instrument source, landing page, engagement and lead events and combine volume with quality. Define a qualified lead: served market, compatible need, company or seniority, consent and genuine intent.
Request a model linking prompts to pages, pages to calls to action, calls to action to events and events to pipeline. When direct attribution is absent, use assisted evidence with explicit limits. Do not invent “AI revenue” through opaque windows or attribution models. The objective is a credible demand path, not calling every referral session a success.
How to score agencies in 24 points
Assign zero to four points to every proof: zero means absent; one, a promise; two, a documented method; three, partial evidence; four, reproducible execution. Require at least three in Eligibility, Source and Measurement. Do not offset weak content with technical work or weak measurement with anecdotal appearances. Evaluate the actual team too: strategy, research, content, technical SEO, development, analytics, external authority and demand generation.
A recommended 8-to-12-week pilot
Select one priority service, three language markets and a small set of commercial questions. Record the baseline, correct blockers, improve existing pages and create a few original assets. Configure measurement, run checks at a fixed cadence and track impressions, citations, referrals and leads. The pilot should finish with learning by prompt and page—not a report of “optimizations” without observable movement.
- Minimum outputs: demand map; technical audit; prioritized backlog; briefs with original contribution; published pages; authority plan; measurement protocol; evidence records; referral and lead dashboard; change history; risks and next tests.
Red flags and next step
- Citation guarantees; mass page creation; llms.txt sold as the main solution; prompts selected after results; appearances without a URL counted as citations; artificial mentions or reviews; traffic called a lead; no baseline; single-platform dependence; content without expertise or sources; dashboard without raw evidence; no conversion plan.
Use Evidence-6 to structure procurement with https://makinai.co/insights/en/how-to-write-rfp-ai-services and compare general capability with https://makinai.co/insights/en/how-to-choose-ai-implementation-company-brazil-scorecard. To connect strategy, content, media, measurement and machine discovery into a growth system, visit https://makinai.co/services/en/digital-marketing-media-performance-growth-agency and https://makinai.co/services/en/data-content-intelligence-systems.