OpenAI is testing an advertising format inside ChatGPT: an ad shown beside the assistant while it generates an image, in a US-first trial with a limited group of advertisers, alongside the measurement and brand-safety tooling an advertising business needs before money moves at scale (OpenAI). The paid half of AI search is now being built. Being named in an AI answer is still unpaid, no bid sits behind it, and the citation sits inside the answer rather than in a list a paid unit can push down the page. Earn it now. That is the one decisive move this edition argues for, and everything below explains why the position holds.
A bid can only exist where a paid slot exists, and the only slot OpenAI is testing sits beside image generation, not beside a text answer. So when ChatGPT, Gemini or Perplexity answer a question and cite the businesses they can read and trust, the citation currently carries no adjacent paid unit at all.
Search ran this arc once, and the early movers still hold the ground
Google launched AdWords in October 2000 on a flat-rate model, and the cost-per-click auction arrived in February 2002 as AdWords Select. From that point, appearing at the top of the page for competitive terms stopped being earned and started being bought, with the price set by whoever bid highest. The businesses that built organic authority before their market got expensive still appear for their terms without paying per click. The latecomers rent the same ground at auction, every click, every month.
AI answers are now acquiring that same machinery, and the detail that matters is how the price gets set. OpenAI's tested format already ships with two distinct systems an ad business depends on: measurement, which counts impressions and reads brand lift, and attribution, which credits conversions back to the placement. AdWords shows the mechanism that turns the first into the second. A flat placement rate works only while slots are plentiful; once demand for a position exceeds the supply of positions, a fixed price cannot clear the market, so the seller switches to an auction and lets the highest bid ration the scarce impressions. Impression share, your impressions as a fraction of those you were eligible for, is simply the measure you read off the back of that rationing, not a quantity the auction hands out. Pricing that path is exactly what the current test exists to do. The format that follows will carry a price. The citation does not yet.
Here is the objection that kills a lazy version of this analogy: organic search did not stay free and visible. Ads and shopping units pushed the free results down the page, so early movers kept their rankings but lost the screen. If AI answers work the same way, earning citation early buys you nothing. The next section is where that breaks.
AI answers cannot bury a citation the way search buried organic
An AI answer has no separate list to push down. In classic search, paid results and free results are different slots on the same page, which is why a stack of paid and shopping units could push the top organic result below the fold. An AI answer is one composed response, and the sources it cites are woven into the substance of what the user reads. A paid unit can sit beside that answer or above it, but it cannot demote the sources the answer is built from, because there is no second list to demote them into. OpenAI's tested format proves the point: the ad runs beside the assistant, not inside the cited answer.
That is why earned citation is structurally more durable than an organic ranking was. The rival view, held by firms telling clients this is simply SEO again and that the ad layer will bury citation exactly as it buried organic, rests on page geometry that does not exist here.
Those same firms will tell you to retrofit readability later at the same cost, so there is no rush. The labour does cost the same whenever you do it. What it buys does not. Done now, a fixed one-off cost earns a citation that recurs for free. Done after the paid phase lands, the identical work only qualifies you to compete for a slot an advertiser can also buy, and appearing shifts from a settled capital cost to a recurring media cost, charged per impression or per click for as long as you want to be seen, with the price set by the highest bidder rather than by you. Same input, a recurring bill in place of a free position.
One honest caveat sets the clock. On the engines we track, this durability holds while they rank citations on usefulness rather than payment. The mechanism that would end it is visible: an engine would have to start seeding paid sources into the citation list itself, rather than beside the answer as OpenAI's current test does. The leading indicator we watch is exactly that boundary, whether a cited source in the body of an answer can be traced to an advertiser relationship. While the ad sits beside the answer and the citations are earned, the advantage holds.
Organic citation is still unpaid, and three fixes earn it
Citation is earned at the retrieval step, not through bidding. ChatGPT search, to take one engine precisely, answers by running a live query, pulling back a set of pages it can read and trust, and composing its reply from them, so your inclusion depends on being retrievable and quotable at the moment the question is asked. Gemini, Perplexity and Google's AI Overviews each reach their sources by their own route, and I am not claiming a single shared technique across them. What every one of them requires is the same: a machine has to be able to read your site, confirm who you are, and repeat your facts without risk. Three fixes do most of the work to make a business safe to name, and we measure each as a sub-score in the Crane Index™:
- Organisation and LocalBusiness schema (schema.org), so a machine does not have to guess who you are. Structured data states your name, type and location in a format engines read directly.
- One consistent identity across every page, the same business name, address and contact details, with no contradictions for a machine to trip over.
- Your trading facts in plain page text, not locked inside images or PDFs, marked up where it helps with Product, Offer and FAQPage schema, so what you sell, where and at what price can be quoted.
Across more than three thousand UK sites scored on the Crane Index™, fewer than one in twenty clear 70 out of 100, and the median company site sits near 18 out of 100 for how clearly a machine can describe it. Most UK businesses are not yet earning the free citation, which means the ground is open. A low score is not a penalty. It is an unclaimed position.
For a leader the move is simple. Treat AI readability as a capital project now, and direct whoever owns your website to make the business machine-readable while a citation still costs work rather than media budget. A business made readable while the ad sits beside the answer carries its naming into whatever the paid layer becomes. One that waits inherits whatever rules the auction writes without it.
