Ads inside ChatGPT can draw on a conversation's context and a user's permitted personalization, but they do not inform the model's answer. Abhilash Edathil, who works on OpenAI's monetization team, made that distinction explicit during an on-demand webinar with Go Fish Digital covering how earned and paid AI visibility fit together.

Edathil demonstrated a travel-planning conversation in which an ad appeared alongside, but apart from, the organic answer. AJ of Go Fish Digital framed the split in plain terms: a brand recommendation inside an answer is earned visibility, while a labeled ad is paid visibility. Two different brands occupying those two positions is not a contradiction.

The distinction matters for planning. A paid test can put an offer in front of someone during a relevant conversation, but it does not buy inclusion in the model's answer. Conversely, an earned mention is not an ad impression and is not proof that the buyer clicked through.

Measure GEO as a pattern, not a ranking

A single prompt response is too variable to function like a traditional rank position, AJ argued. He proposed a three-part framework:

• Presence: is the brand mentioned or cited for a buyer question?
• Representation: is that description accurate and current?
• Competitiveness: how often does it appear relative to relevant rivals?

To make the assessment useful, AJ suggested starting with roughly 20–40 questions customers actually ask, running them repeatedly over a week or two, and keeping the model and settings consistent. Marketers should look for recurring gaps rather than treating one answer as a verdict.

Patrick Algrim added that a brand should examine the quality of a recommendation, not merely count mentions. The evidence available about its products and services shapes what the system can explain.

Fix the evidence on your site first

Asked whether to prioritize a company website or off-site mentions, both Go Fish Digital speakers started with the site. Algrim used a moving-company example: if a company offers cross-country moves but never says so on its site and has no corroborating evidence elsewhere, it should not assume an AI system will infer that service. Its own pages need to answer buyer questions, specify what the company actually does, and stay consistent with external descriptions.

Algrim explained why marketers should start there (33:44):

So start with your website because you can control it. Make sure everything is, you know, again, factually true, connected to a cohesive story about your brand, product, service, what it is that you offer, and, and really just start there.

AJ added a technical first check: make sure the relevant pages are accessible to search crawlers rather than blocked by robots rules or a CDN. Reviews, PR, and other external evidence still matter, but they cannot rescue an unclear or contradictory account of the business.

Test ads against an objective, treat AI referrals as partial data

In the audience Q&A, Edathil said ChatGPT ads were shown to eligible adults in the Free and Go versions, not paid versions, and that availability by market and vertical was evolving. He pointed advertisers to OpenAI's ads manager to check current eligibility and start a test. For measurement, he recommended choosing the desired outcome first — reach, traffic, or conversions — connecting conversion data through the available pixel or API when appropriate, and changing creative, bids, or budgets based on results. He described the platform as it stood at the time of the webinar, not as a guarantee of availability or performance for every advertiser.

AJ warned that AI influence can disappear from last-click reporting. A person may discover a brand in a chatbot, search for it separately, or paste a URL into a browser. GA4's AI-assistant referrals therefore show only part of the journey. He suggested supplementing them with customer self-reporting and trends in branded and direct traffic, while keeping assistants separate in reports. Algrim recommended testing specific website or campaign changes over time rather than assigning every increase to a single source without evidence.

What to do next

The webinar's practical takeaway: choose one buyer problem and test a focused improvement, rather than chasing every new GEO tactic. A useful first pass looks like this:

• Write down the buyer questions that matter and check recurring answers for presence, accuracy, and competitor comparisons under consistent settings.
• Audit the pages that should answer those questions: clarify services, eligibility, and differentiators, then check crawler access and conflicting off-site claims.
• If paid placement fits the objective and the account is eligible, run a limited ChatGPT ads test with defined outcomes and conversion measurement.
• Review AI referral data alongside branded demand, direct traffic, and customer-reported influence, and document changes so later results have context.

For marketers weighing budgets between paid AI placements and GEO investment, the webinar's answer is structural: the two channels do different jobs, and neither substitutes for a website that states plainly what the business sells.