The AI Shelf: What Decides Which Products an Answer Engine Recommends

Ask ChatGPT about your company by name and it gets the answer right almost every time. Ask it what to buy in your category and your brand disappears. Across 2,240 queries covering 112 companies, ChatGPT identified brands correctly 99.4 percent of the time when asked directly and surfaced them in open discovery questions 3.32 percent of the time. Perplexity showed the same collapse.

That gap is where a new shelf is being stocked. For a century the retail shelf decided what got bought, and space on it was purchasable. The new shelf is assembled at the moment a buyer asks a question, and nobody sells space on it.

AEO and GEO, Graded

Answer engine optimization and generative engine optimization are the two names the market uses for the same problem: getting an AI system to cite and recommend you. The advice sold under both labels is largely untested. The AI Shelf grades every substantive finding from A to D, names which studies were published by companies selling the remedy, and lists the widely repeated figures that failed source verification. Nine were excluded during research, including a claim about FAQ markup that no primary source supports.

The paper is built for the commerce leader who has to approve or decline an AEO or GEO program this quarter and wants to know what the evidence supports before signing.

Two Layers, and Only One is Managed

What decides the outcome divides cleanly in two. The Mention Layer is the crawled open web: forums, reviews, video, editorial coverage. You do not own it. The Merchant Layer is your structured product feed. You own it completely, and in most enterprises it reports to the paid media team and is tuned for bid efficiency.

The paper measures which of the two wins the first product slot on AI shopping surfaces. The answer is not the one your content budget is currently funding, and it changes who in your organization should own this work.

One Check You Can Run This Week

Ask your infrastructure team what your CDN and firewall actually return to each named AI crawler. Not what robots.txt says. What the servers answer when the crawler arrives. A significant share of large enterprise sites publish an open policy and return HTTP 403 in practice, and nothing in your analytics will tell you. The paper has the measured failure rate, the full list of agents to test, and the distinction between training crawlers and retrieval agents that makes blocking decisions expensive in both directions.

Why This is the Only AEO or GEO Document You Need

Most AEO and GEO guidance covers content tactics and stops there. It leaves out the product feed, which is where a commerce recommendation actually gets decided. It also skips the prior question of whether the tactics work at all.

The AI Shelf covers both layers and every engine with a commerce footprint. We read the 55 sources so your team does not have to, graded each one A to D, and named which were published by companies selling the remedy. Nine circulating figures were thrown out for failing verification. Where nobody has measured something, the paper says so and marks the gap.

Read it once and you have the field. You will not need the next twelve vendor blog posts.

What You Get in This White Paper

The AI Shelf is built as a reference document. It is meant to be opened during a vendor call and used to argue for a budget.

Frameworks and Reference Tables

• The Legibility Stack: Four layers ordered by evidence quality and cost, each with a named owner, an effort estimate, and the grade of the evidence behind it.
• The factor ladder: Fifteen visibility factors ranked, each with its best measured effect, its evidence type, and the watch-out that vendor decks leave out.
• The engine matrix: How ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Perplexity, Copilot, Claude, and Amazon's assistants each retrieve, what each one cites, and the first three moves for each.
• The factor and engine grid: Twenty-two optimization parameters scored across seven engines. It separates the hard gates from the tactics that were tested and found to do nothing. Untested cells are marked untested instead of filled in.
• Feed requirements by platform: What Google, OpenAI, Perplexity, and Amazon each demand, with the identifier and accuracy gates that silently remove products when they fail.

First moves by company type for retailers, consumer brands, marketplace sellers, and industrial manufacturers.

55 graded sources and a glossary of the terms your vendors use without defining them.

Questions it Answers

• Which layer wins the first product slot on AI shopping surfaces, and why most enterprises have it pointed at the wrong team
• Whether llms.txt, schema markup, FAQ blocks, and answer formatting move AI citations at all, with the controlled tests that checked
• Why the two largest search operators publicly disagree on whether AEO and GEO exist as disciplines
• How many repetitions your measurement needs before the number means anything, and why every ranking dashboard falls short of it
• Which engines admit the most spam into their answers, and by what multiple
• Where the incumbency advantage in AI recommendations breaks, and how small the quality edge has to be to break it
• What transfers to industrial and B2B commerce, and the open question there that nobody has measured

Actions You Can Take Now

• Six actions to start with, ordered by evidence quality and inverse cost. Four of them are free.
• A ten-item priority list covering everything in the paper, in execution order, with the reason each item sits where it does.
• Three programs not to fund this year, with the public record that explains why.
• Six questions to put to any vendor selling AI visibility. Nobody in this market has a published answer to the sixth.
• Two free diagnostic checks that tell you which layer is broken, and which team owns the fix, before you spend anything.

Who it is For

Commerce leaders at retailers, consumer brands, direct-to-consumer businesses, marketplace sellers, and manufacturers selling through distribution. If you publish structured records describing things you sell, the evidence applies to you, whether those records describe breakfast cereal or bearing assemblies.

Download the white paper for the full evidence base on AI search visibility, or talk to our commerce and product experience teams directly at www.gspann.com/contact-us. GSPANN runs this work on the platforms most enterprise catalogs already use: Adobe Commerce, Salesforce Commerce Cloud, and commercetools.