AEO 101 for D2C brands: how answer engines find, choose and read your products

When a shopper asks ChatGPT or Google's AI Overviews what to buy, the answer feels like it comes from everything the model knows. It does not. For a shopping question, an answer engine goes out, fetches a handful of pages, lifts a few passages from them and writes its answer from those passages alone. If your product page was not in that handful, it does not matter how good the product or the page is.

This is our D2C version of Lesson 2 of Profound 101, in which Josh Blyskal of Profound walks one prompt through an answer engine and turns it into a three-gate diagnostic. We have rebuilt the lesson around a D2C shopping prompt, then added what we see when we audit Shopify stores: the robots.txt rules, product page titles, accordions and app-loaded widgets that quietly decide whether a brand gets cited.

The one rule to remember

An answer engine can only answer from what it fetched. Not from what is true about your products, and not from what exists somewhere on your site, but from the pages and passages it actually pulled into its context while writing the answer.

Profound describes an answer engine as a language model chained to a browser: the model is the brain that reads and writes, the browser is the hands that go and fetch. Every time the hands fetch, the system makes a choice under a constraint, and every choice is a place where your store can drop out.

Follow one shopping prompt through the engine

Here is a prompt of the kind we test in Eldor audits for jewellery brands:

"I want gold hoops I can wear every day, even in the shower. Under ₹2,000, and okay for sensitive ears. Which brands should I look at, and can I return them if they don't suit me?"

Step 1: the prompt becomes several searches

The first constraint is that the engine cannot search that whole paragraph as typed. So it translates it into a handful of short, search-shaped queries. This is query fan-out. Our prompt might become:

• best waterproof gold hoop earrings

• gold plated hoops under ₹2,000 India

• hypoallergenic earrings for sensitive ears

• demi-fine jewellery brands India

• [brand] return policy

Each of those searches comes back with its own list of possible sources. Your product does not compete on the shopper's sentence. It competes on each sub-query separately, and a store can win one list and be absent from the other four.

Step 2: the engine judges each page by its cover

The second constraint is that the engine cannot open every result. Before it clicks, it looks at what Profound calls the cover, or the billboard: the URL, the title, the snippet and sometimes a last-updated date. That is all it has to decide whether a page probably contains the answer it needs.

It picks a few pages, not one, because it wants enough evidence to write a confident answer. In the lesson's words, you are not just writing good content, you are "marketing the cover of the book to a machine."

Step 3: the engine extracts chunks, it does not read

The third constraint is that the engine cannot read your whole page the way a shopper does. It does not scroll, look at the photos and absorb the brand. It lifts small, self-contained blocks of text and brings those back. If the answer is buried, scattered across the page or hidden behind the interface, extraction gets harder, and a cleaner page from someone else gets used instead.

Step 4: the answer is written from those chunks

Finally the model synthesises its answer from the chunks it collected, and the citations are simply the engine pointing at the places it fetched while writing. That is the full loop: three constraints, three decisions. It cannot search your whole prompt, it cannot open the whole web and it cannot read the whole page.

The three gates, applied to a D2C store

The three constraints give you a diagnostic. When your brand does not show up for a prompt it should win, ask which gate it failed: was the page not fetchable, not chosen, or not extractable? Each gate has different causes and different fixes, so naming the gate stops the guesswork.

Gate 1: fetchable. Can the engine reach and read the page at all?

The symptom is total invisibility: the engine never mentions you, even for branded prompts. This is not a content problem. Common causes we find on D2C stores:

• AI search crawlers blocked in robots.txt. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. Blocking GPTBot, which is used for model training, is a separate decision and does not remove you from ChatGPT search. Check that OAI-SearchBot, PerplexityBot and Bingbot are allowed.

• Bot protection at the CDN. Cloudflare started asking new domains in July 2025 whether to allow AI crawlers, and existing sites can block them in one click. If your store sits behind Cloudflare or another firewall, check what it does to AI crawlers, not just to Googlebot.

• Pages that are not on the open web. A store still behind Shopify's password page, a product that is unpublished from the online store channel, a members-only collection, or a policy that only lives inside a PDF or a login. If a shopper needs a password, so does the crawler.

• Noindex left on by accident. Apps and theme settings sometimes add noindex to collection, search or landing pages. Google's AI Overviews draw from the pages Googlebot indexes, so a noindexed page cannot be used there.

• Google-Extended is not the switch for AI Overviews. Google says the Google-Extended token does not affect inclusion in Google Search. Blocking it does not hide you from AI Overviews, and allowing it does not get you in. Indexing does.

Fix fetchability first. Nothing in the next two gates matters for a page the engine cannot load.

Gate 2: chosen. Does your cover look like the answer?

The symptom: your pages are indexed, you may even rank, but the engine never cites you. Usually you are not missing from the internet. You just do not look like you fit the job.

Profound's example is a running shoe page titled around one shoe, for a prompt asking for the best shoes. One shoe and one angle, when the shopper asked for coverage and comparison. The D2C version of this is everywhere:

• Product titles that are only names. A product page titled "Aurora Hoops" tells the engine nothing for "waterproof gold hoops for daily wear". A title like "Aurora 18k gold-plated waterproof hoops for daily wear" puts the material, the use case and the shopper's words on the cover.

• Snippets that sell instead of answer. "Shop the latest collection now" is not evidence. A description that states what the product is, who it suits, the price band and the key fact (waterproof, hypoallergenic, 15-day returns) gives the engine a reason to click.

• No page shaped like the question. "Best" and "vs" prompts want options. If your only pages are single products, you are the one-shoe review. A buying guide or a well-built collection page, organised by need (everyday, sensitive ears, under ₹2,000, gifting), is the answer-shaped cover for those searches.

• Stale signals. Some engines see a last-updated date on the cover. A buying guide last touched two years ago, quoting old prices, is an easy page to pass over.

This is the gate where most indexed D2C stores lose. The fix is not more content. It is giving each important page a cover that clearly matches one sub-query in the fan-out.

Gate 3: extractable. Can the engine lift a clean chunk?

The symptom: the engine opens your page, sometimes even cites it, but the answer it gives is vague or wrong, or a competitor's fact gets used instead of yours. The answer never made it into the engine's context. On product pages, the usual causes are:

• Facts locked in images. Size charts, fabric details, ingredient lists and "what's in the box" uploaded as images. Retrieval is text first. If the only place your sizing lives is a JPEG, most systems cannot use it.

• Content that only appears after JavaScript runs. Vercel's study of AI crawler traffic found that the major AI crawlers, such as GPTBot and ClaudeBot, fetch JavaScript files but do not execute them. Reviews, Q and A, size guides and delivery estimates injected by apps after the page loads may simply not exist for them. View the raw HTML of a product page and search for the fact you want quoted.

• Answers buried in accordions or tabs. Many Shopify themes render collapsible tabs into the page HTML, which crawlers can usually read. The risk is a tab whose content an app loads only when it is clicked. Either way, the facts shoppers ask about most (material, fit, care, delivery time, returns) should also appear as plain text near the top of the description.

• Long brand-story intros. Three paragraphs about your founder's grandmother before the page says "18k gold plating on 316L stainless steel, waterproof, nickel-free" makes the useful chunk harder to find and easier to skip.

• Scattered facts. Price in one widget, stock in another, the returns window in the footer and COD terms on a separate page. A self-contained block that answers one question in two or three sentences is the unit an engine can lift.

• Video without text. A styling video or an unboxing reel with no transcript or caption copy gives a shopper something and an engine nothing.

Feeds are a second front door

Pages are not the only thing engines fetch for shopping questions. Google's shopping answers draw on its Shopping Graph, which is built largely from Merchant Center feeds, and ChatGPT shopping reads merchant product feeds. The same three gates apply to your feed. Is it submitted and approved (fetchable)? Do product titles carry the attributes shoppers search for (chosen)? Are price, stock, material and variants filled in as clean fields rather than left blank (extractable)? We covered feeds and the rest of the pipeline in How AI answer engines pick products: AEO for D2C stores.

Diagnose a missing product in ten seconds

When a product should be in an answer and is not, match the symptom to the gate before you change anything:

• Not mentioned anywhere, even for your brand name: gate 1, fetchable. Check robots.txt, your CDN's bot settings, noindex tags, the password page and whether the page is published.

• Indexed, maybe ranking, but never cited: gate 2, chosen. Rewrite the title and description so the cover matches the sub-query, and build the buying guide or collection page that "best" prompts need.

• Cited, but the answer is wrong or vague: gate 3, extractable. Move key facts out of images, apps and accordions into plain text, and put a short answer block near the top.

• Right on your site, wrong in the answer: the engine fetched someone else's page, such as a reseller listing with an old price or a marketplace page with the wrong stock. That is also a gate 2 problem: their cover won the click. Fix your own page's cover, and correct or update the third-party listing where you can.

A two-week plan for a lean D2C team

• Days 1 to 2, baseline. Write 25 real shopper prompts for your category and price band, including "best", "vs", budget and policy questions. Run them in ChatGPT and Google, and note for each one whether you are named, linked or absent, and which sources are cited.

• Days 3 to 4, gate 1. Audit robots.txt, CDN bot rules and noindex tags. Confirm your top 20 products and your policy pages load for anyone without a login.

• Days 5 to 9, gate 2. Rewrite titles and meta descriptions for your top 20 products with material, use case and shopper language. Build or rework one buying guide or collection page for your most valuable "best" prompt.

• Days 10 to 13, gate 3. Move size charts, materials and care into text. Add a short answer block at the top of each top product page. Check that reviews and delivery estimates exist in the raw HTML.

• Day 14, re-run. Run the same 25 prompts again. AI answers move from week to week, so keep the prompt set fixed and judge the trend over several weekly runs, not one answer.

Where Eldor fits

This diagnostic is what the free Eldor AI audit automates for D2C stores. We run 25 shopper prompts written for your category and price band on ChatGPT and Google AI Overviews, check 100 of your products against your live store, and show which prompts you win, lose or are absent from, with the competitor and source behind every lost answer and a fix plan ranked by impact.

On a paid plan, Eldor tracks 100 or 300 prompts every week, flags every place an engine quotes the wrong price, stock or policy, and fixes the product data at the source with enrichment and page rewrites, so the next time an engine fetches your page, the chunk it lifts is the right one.

FAQ

What is AEO?

AEO, or answer engine optimisation, is the practice of making your brand and products the answer AI engines give when shoppers ask buying questions. For a D2C store it covers being fetchable by AI crawlers, having pages and feeds that look like the answer, and stating product facts in text an engine can lift.

Is AEO different from SEO?

They overlap at gate 1, because both need crawlable, indexed pages. They diverge after that. SEO wins a ranking position for a query. An answer engine runs several queries for one prompt, opens only a few results based on their cover and writes from extracted chunks, so page titles, answer-shaped pages and plain-text facts matter more than they did for blue links.

Should a Shopify store block GPTBot?

That is a business decision about model training. It does not decide ChatGPT search visibility. OpenAI says opting out of OAI-SearchBot keeps a site out of ChatGPT search answers, so if you want to be recommended there, allow OAI-SearchBot.

Why does ChatGPT quote the wrong price for my product?

Usually because it fetched a page that is not yours, such as a reseller or marketplace listing, or an old cached version of your own page. Check the sources the engine cites, fix your own product page so its cover and price are clear, and update third-party listings where you can.

How long until changes show up in AI answers?

It varies by engine and page. OpenAI notes that robots.txt changes can take about 24 hours to affect ChatGPT search, while content changes depend on when a page is next crawled. Track a fixed set of prompts weekly and judge the trend over several runs.

Sources: Josh Blyskal, Profound, "Profound 101: How Answer Engines Retrieve Information" (Lesson 2) and its course notes; OpenAI, "Overview of OpenAI crawlers"; Google Search Central, Google crawler documentation on Google-Extended; Vercel, "The rise of the AI crawler" (December 2024); MIT Technology Review on Cloudflare's AI crawler controls (July 2025).