aisearchoptimization.hu / Articles / Checklist

GEO Vendor Due Diligence: 5 Questions to Vet AI Visibility Agencies

A practical checklist to separate evidence-based AI visibility work from marketing hype and ranking guarantees.

By Zoltán Varga2026-11-057 min read

What Exactly Will Be Measured and How Often

Ask the vendor to define the specific metrics they track for Generative Engine Optimization. A single snapshot of your brand appearing once in a chat response is not a valid metric. Reliable measurement requires running hundreds of prompts across multiple models like ChatGPT, Perplexity, and Gemini to calculate a statistical visibility rate. This approach accounts for the inherent randomness in how large language models retrieve and present information.

Vendors who offer a one-time score without explaining their sampling method are likely selling a vanity metric. Research from 2026 indicates that the probability of two identical AI responses listing the same brand is often below one percent. A credible provider will explain that they measure the frequency of appearance over time, not a fixed position. They should also clarify if they track citation accuracy or just brand mention.

Your contract must specify the frequency of reporting. Monthly reports based on a single day of testing are insufficient to prove progress. Look for a partner who runs tests weekly or daily and aggregates the data to show trends. This consistency helps distinguish between algorithmic noise and genuine improvements in your content authority. Without this volume of data, you cannot verify if their actions caused the change.

Ensure the vendor measures performance across different query types. Some agencies only test generic keywords while ignoring long-tail questions where your business actually competes. A robust audit checks how your brand appears when users ask for comparisons, solutions, or specific industry insights. The goal is to see your content cited as a solution, not just mentioned in passing. This distinction separates true visibility from accidental exposure.

What Evidence Backs Their Method

Request case studies that show before-and-after data from repeated testing cycles. Avoid vendors who share screenshots of a single chat window as proof of success. Real evidence comes from aggregated data showing a rise in citation frequency over a defined period. The vendor should explain the specific changes made to your content structure and how those changes correlated with the data shift.

Check if the vendor references established research on how LLMs process information. Methods based on the concept of 'priming' or 'answer priming' suggest that LLMs scan content non-linearly. Effective strategies involve creating summary blocks and clear hierarchies that match how AI models extract facts. Ask if their team understands the difference between optimizing for search engines and optimizing for generative models.

Verify that the vendor does not rely on proprietary black-box tools without explaining the underlying logic. Transparency is key. They should be able to articulate why certain schema markup or content formats improve visibility. If they claim a secret algorithm that no one else knows, treat this as a red flag. Valid GEO work relies on public principles of entity recognition and structured data, not hidden tricks.

Look for evidence that the vendor understands the limitations of current AI models. Most large language models cannot access paid subscription databases and rely on open-access resources. A competent agency will focus on making your public content the most authoritative source available. They should not promise to get you into closed datasets or private knowledge bases that are inaccessible to the general public.

Ask about their approach to authorial authority. Building natural citations requires establishing your brand as a trusted source over time. This involves consistent publishing, clear authorship, and factual accuracy. Vendors who skip this step and try to force citations through technical hacks are setting you up for failure. Long-term visibility depends on reputation, not just code.

Red Flags: Guarantees and Ranking Promises

Reject any agency that guarantees a specific ranking or a fixed number of citations. Large language models do not have a static ranking system like traditional search engines. Their outputs vary based on the prompt, the model version, and the current context. Promising a guaranteed spot in an AI answer is a sign of dishonesty or a fundamental misunderstanding of the technology.

Be wary of vendors who use terms like 'guaranteed' or 'certified' in their marketing materials. No certification exists for AI visibility because the field is too new and dynamic. The technology evolves rapidly, and methods that work today may fail tomorrow. A professional vendor will discuss probabilities and trends rather than absolute promises. They should manage your expectations realistically.

Watch out for agencies that promise immediate results within days. Building authority takes time. Even technical fixes like correcting schema errors or fixing broken links require crawling and indexing cycles. If a vendor claims they can get you cited in ChatGPT within 48 hours, they are likely using manipulative tactics that could harm your reputation. Sustainable growth follows a slower, more deliberate path.

Question any proposal that involves buying mentions or creating fake reviews. These practices violate the terms of service of most platforms and can lead to penalties. AI models are trained to detect patterns of manipulation and may suppress content that appears inauthentic. Focus on organic growth strategies that build genuine trust with your audience and the algorithms that serve them.

Avoid vendors who promise to manipulate the model's internal weights or training data. This is technically impossible for external parties. You cannot change how an LLM learns or stores knowledge. The only lever you have is the quality and structure of the content you publish. Any claim to the contrary is a sales tactic designed to exploit your lack of technical knowledge.

Manipulation Risk and Durability

Ask how the vendor handles the risk of manipulation detection. Some tactics, such as stuffing content with keywords or creating artificial links, can trigger spam filters. AI models are becoming better at identifying low-quality or deceptive content. A safe strategy prioritizes clarity, accuracy, and natural language over keyword density. This ensures your content remains visible even as models evolve.

Inquire about the durability of their methods. Techniques that rely on exploiting a specific bug in a model are short-lived. Once the bug is patched, your visibility will drop. Sustainable strategies focus on building a strong foundation of content authority. This includes consistent publishing, clear entity definitions, and high-quality data structures. These elements remain valuable regardless of model updates.

Discuss the risk of 'context rot' or information decay. As models process more data, older information can become less relevant. A good vendor will have a plan for updating your content regularly. They should monitor your visibility and suggest revisions to keep your information current. Static content may eventually be ignored by newer, more efficient models.

Consider the risk of relying on a single platform. If you optimize only for ChatGPT, you miss opportunities in Perplexity or Gemini. Diversify your strategy to cover multiple models. This reduces the risk of being affected by changes in one specific system. A robust approach treats AI visibility as a multi-channel effort, similar to traditional search engine optimization.

Ask about their policy on ethical guidelines. Reputable vendors adhere to industry standards and avoid gray-hat tactics. They should be transparent about the risks involved in any proposed strategy. If they hesitate to discuss ethics or suggest cutting corners, walk away. Your brand reputation is too valuable to risk on unproven or dangerous methods.

Reporting Standards and Exit Terms

Define clear reporting standards in your contract. You should receive regular updates that include raw data, not just charts. The report should show the number of prompts run, the models tested, and the specific queries used. This transparency allows you to verify the results independently. Without access to the underlying data, you cannot trust the conclusions drawn by the vendor.

Establish exit terms that protect your intellectual property and data. Ensure you retain full ownership of all content created during the engagement. The vendor should provide all raw data and analysis files upon termination. This prevents them from holding your data hostage if you decide to switch providers. Clear exit clauses are essential for maintaining control over your digital assets.

Ask about the transition plan if you end the relationship. A good vendor will help you understand the current state of your visibility and suggest next steps. They should not leave you with a broken strategy or missing documentation. The handover should include a summary of what worked, what didn't, and recommendations for future improvements. This ensures continuity even after they leave.

Verify that the vendor uses secure data handling practices. Your business data and customer information must be protected. Ask about their security protocols and compliance with data privacy regulations. If they cannot provide details on how they store and transmit data, consider it a significant risk. Security is a non-negotiable aspect of any professional partnership.

Review the pricing model carefully. Avoid contracts that lock you into long-term commitments without performance milestones. Prefer a model that allows you to pause or terminate the agreement if results are not met. This flexibility protects your budget and ensures you are paying for value. Transparent pricing builds trust and aligns the interests of both parties.

AI-assisted: drafted with a language model from the author's research notes, then edited and checked before publication. Author: Zoltán Varga, AI visibility researcher and AI ecosystem architect, Budapest.