# Red Flags in GEO and AEO Agency Proposals > The biggest GEO and AEO agency red flags are not a lack of AI jargon. They are guarantees, untestable methods, missing evidence, weak approval controls and proposals that cannot explain what the agency will actually change. Canonical page: [https://narattivo.com/insights/geo-agency-red-flags/](https://narattivo.com/insights/geo-agency-red-flags/) Last updated: 2026-10-18 A GEO or AEO agency proposal deserves a pause before signing if it guarantees AI citations, rankings or recommendations; cannot identify the pages and facts it will improve; or hides delivery behind phrases such as “optimise for every answer engine.” A credible proposal sells a defined programme of research, content improvement, technical implementation and measurement. It does not sell control over a future answer from an AI assistant. That distinction matters commercially. Generative engine optimisation (GEO) and answer engine optimisation (AEO) can describe useful work: making a company’s information clearer, better evidenced and easier to maintain. They can also be used to repackage ordinary content or SEO work with an outcome nobody can command. Google’s guidance is a useful reality check. Its [SEO Starter Guide](https://developers.google.com/search/docs/fundamentals/seo-starter-guide) says there are no secrets that automatically place a site first, no guarantee that a page will be included, and changes may take time or produce no noticeable effect. Its systems are intended to prioritise helpful, reliable, people-first information rather than efforts designed to manipulate rankings. Judge an agency proposal against that standard, not against the confidence of its sales language. ## Proposal claims that should trigger a hard question Proposal claimWhy it is riskyEvidence to requestA credible replacement commitment “We guarantee citations or brand mentions in ChatGPT, Gemini or Perplexity.”No agency controls a named assistant’s future source selection or wording. The promise turns an uncertain outcome into a deliverable.Ask which work is guaranteed, which outcomes are only observed, and how reporting distinguishes the two.“We will improve named pages, factual support and monitoring, then report what we observe without guaranteeing mentions.” “We optimise for every answer engine.”This does not identify platforms, assets, method, limits, monitoring or approval requirements.Request a prioritised platform list, workstreams, monitoring method and exclusions.“We will assess relevant search and AI-assisted discovery surfaces and prioritise assets we can substantively improve.” “Our proprietary GEO framework handles everything.”A framework may be useful, but secrecy is not a scope. You still need to know what changes will be made to your site.Ask for an anonymised change log, sample delivery ledger and acceptance criteria.A named audit, content plan, technical ticket list, implementation process and monthly decision record. “Schema will get you into AI answers.”Structured data can clarify page meaning for search engines, but it is not a switch for AI-answer inclusion.Ask which pages and factual fields are involved, how claims will be checked and how markup will be validated.Accurate, page-specific markup alongside stronger visible facts, sources and technical maintenance. “We will create 100 AI-ready articles.”Volume does not prove usefulness. It can create thin, repetitive or unsupported pages your team must later repair.Request topic rationale, source standards, subject-matter review, update ownership and a sample brief.A smaller prioritised page set tied to buyer questions, evidence gaps and commercial intent. “We will report visibility.”Visibility is too vague to manage. A polished dashboard can hide missing work, an absent baseline or unclear data limits.Ask for the starting inventory, tracked questions, collection limits and decisions the report will support.A report showing changes made, evidence added, observed signals, unresolved risks and the next test. “We only need publishing access.”Uncontrolled changes can introduce inaccurate claims, stale prices, compliance issues or technical damage.Request roles, approval stages, staging rules, rollback process and ownership of credentials.Defined access with client approval for factual, regulated, brand-sensitive and technical changes. Not every ambitious phrase means an agency is incapable. A salesperson may have compressed a sensible process into poor wording. Give the agency one opportunity to restate the commitment in specific, testable terms. If it cannot or will not do so, the problem is likely structural. ## Buy the work the agency can control A sound proposal separates delivery from outcome. The delivery column belongs in the statement of work; the outcome column belongs in a limitations section and in reporting. This prevents a supplier from being paid mainly for a result that depends on changing systems, competing sources and user context. ### Work an agency can reasonably own - Auditing priority service, product, category, location, comparison and support pages. - Finding missing facts, contradictory claims, weak explanations and statements without adequate support. - Creating or revising briefs with named source, review and update requirements. - Improving page structure so definitions, eligibility, process, conditions and limitations are easy to find. - Implementing accurate structured data where it matches visible content on the relevant page. - Raising technical tickets, recording owners and documenting whether they were implemented. - Maintaining an agreed question set and recording observations over time. ### Outcomes an agency should not promise - A citation, mention or recommendation by a named AI assistant. - A particular ranking, answer placement or index inclusion. - Identical results across tools, prompts, locations, languages or users. - An immediate revenue effect from pages that still depend on trust, sales follow-up or operational capacity. This is not an argument against GEO or AEO. It is an argument for buying them properly. Google explains that [structured data](https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data) gives it explicit clues about a page’s meaning and may make a page eligible for richer search results. That is useful implementation work. It becomes misleading when sold as a mechanism that forces AI visibility. The same principle applies to content. Do not approve a page because it sounds “AI-optimised.” Approve it because it answers a genuine buyer question accurately, adds useful original explanation where possible, and is supported by appropriate evidence. These are closely aligned with Google’s [people-first content guidance](https://developers.google.com/search/docs/fundamentals/creating-helpful-content). ## What a credible GEO or AEO proposal should specify A useful proposal can become a working plan without another vague discovery call. First, it should name the business problem. That might be weak explanations of a B2B service, inconsistent location information, poor evidence for high-intent questions, or a technical backlog affecting important pages. Next, it should define asset scope. “Website optimisation” is not enough. The agency should list, or explain how it will select, service pages, product or category pages, pricing and eligibility pages, knowledge-base articles, author or expert pages, location pages and comparison content. If it cannot identify likely assets, it cannot estimate effort honestly. Insist on acceptance criteria for each workstream. A content deliverable might require an approved brief, a client-verified factual source list, subject-matter review, final copy and a published URL. A technical item might require a ticket, implementation owner, validation record and rollback route. This protects both sides: the agency is assessed on agreed work, while the client is clear about the approvals it owes. ## Set a 30/60/90-day delivery ledger Do not accept a calendar made only of phases such as “optimise,” “amplify” and “dominate.” Ask for a ledger that procurement and marketing can review each month. - First 30 days: baseline asset inventory, priority question set, page and factual audit, source gaps, access map, approval workflow and technical tickets. - By 60 days: priority pages changed or ready for approval, evidence incorporated, appropriate structured-data tasks completed, and a documented list of blocked work. - By 90 days: change log, observations from the agreed monitoring method, lessons from completed work, pages needing updates and the next prioritised test. The ledger should not promise a mention count. Its job is to make inactivity visible and to record client-side delays fairly. If legal review, pricing confirmation or product approval is outstanding for three weeks, that should appear in the record rather than being mistaken for agency underdelivery. ## Apply UAE controls before localising the plan A UAE business should not accept a global template with “Dubai” inserted into the introduction. The proposal should state market coverage and identify the evidence required for local claims. A company serving Dubai, Abu Dhabi and Sharjah may have different service availability, response times, licensing arrangements, branches, languages, prices or customer journeys. If those differences are real, one generic location page can mislead buyers. Ask the agency to complete this scope worksheet before work begins: - Emirate and service area: Which emirates, cities or delivery zones are in scope? Which claims apply everywhere, and which are location-specific? - Language: Will the work cover English, Arabic or both? Who confirms that each version preserves factual meaning rather than merely translating keywords? - Local evidence: Which client-approved records support address, availability, eligibility, process, pricing and contact claims? What will be removed or qualified if it cannot be supported? - Regulated and sensitive claims: Which pages require legal, compliance, clinical, financial or technical review? Name the client approver and expected turnaround. - Ownership: Who supplies facts, approves copy, implements development tickets and maintains pages after the engagement ends? For regulated, high-consideration or locally delivered services, an agency should not write around a lack of proof with broader language. It should flag the gap, request confirmation and either narrow the claim or leave it out. ## Make the final decision with evidence, not hype Score each proposal from 0 to 2: 0 means absent or unconvincing, 1 means partly defined, and 2 means specific and evidenced. Assess the business objective, named asset scope, research standards, content quality control, technical implementation, baseline and monitoring, UAE scope, approval workflow, reporting ledger, access security, exit terms and limitations. A proposal with a lower headline promise but a stronger score is usually the safer purchase. Use the scorecard to force plain answers to these questions: Which pages will you change first, and why? What will we receive in 30, 60 and 90 days? What factual evidence must we provide? Which results are observations rather than commitments? Who owns briefs, research, tickets and accounts at the end? How will you hand over change logs and unresolved recommendations? The right agency may say “we do not know yet” during discovery. That is acceptable if it explains how the answer will be found, who owns it and when it will be documented. What is not acceptable is certainty without a method. Before selecting a partner, use Narattivo’s [10 questions to ask before hiring an AI search optimization agency](https://narattivo.com/insights/questions-to-ask-ai-search-agency/) to turn competing proposals into an evidence-led procurement decision.