What are the best practices for AI recommendation positioning?

Five mechanics. (1) robots.txt allows GPTBot and OAI-SearchBot. (2) Product, Offer, and AggregateRating schema on every advertised PDP. (3) Answer-first PDP copy, the first sentence is the literal answer to the buyer's question. (4) Citation density in publications the model already trusts (Modern Retail, Digiday, Adweek, Reuters, founder interviews on operator podcasts). (5) A multi-engine query harness that measures where you actually appear, not where you hope to.

How do top Shopify brands get AI recommendations?

Their robots.txt allows the AI crawlers. Their PDPs answer the buyer question in the first sentence, not after three paragraphs of brand story. They have at least one mention in a primary publication ChatGPT already cites. Brands the model never names usually miss at least one of those three.

Why are competitors ranking higher in AI recommendations?

The model has more material to pull from when it talks about them. Schema it can convert into a structured answer. Coverage in publications it trusts. A first sentence that matches the question being asked. Each gap is small. Stacked, they decide whether the buyer reads your name or your competitor's.

Sacha Lefebvre Founder, Paid Ai Search · LinkedIn

Best Practices for AI Recommendation Positioning

A buyer asks ChatGPT for a product. The model returns one brand, maybe two. The brand it names wins the sale. The brand it doesn’t find out three months later from a flat revenue line.

Five mechanics decide which brand the model names.

1. Crawl access

The model cannot recommend a brand whose PDPs it cannot read. robots.txt is the front door. Block GPTBot, OAI-SearchBot, or ClaudeBot and the model has no first-party signal, falling back to whatever third-party coverage exists. Often that’s nothing.

Verify in 30 seconds by fetching a competitor’s robots.txt. As of 2026-05-13, allbirds.com/robots.txt and gymshark.com/robots.txt both explicitly allow GPTBot and OAI-SearchBot. The brands the model recommends in DTC footwear and apparel have the front door unlocked.

Fix on Shopify by editing robots.txt.liquid. Allow GPTBot, OAI-SearchBot, ClaudeBot, Google-Extended, PerplexityBot.

2. Schema on every advertised PDP

The model converts unstructured HTML into structured answers. Pre-existing structure beats a guess.

Three schema types do the work on a Shopify PDP. Product (what the page is about). Offer (price, currency, availability). AggregateRating (third-party validation). Shopify ships partial Product schema by default. Offer is usually present but inconsistent on price and availability. AggregateRating is frequently injected client-side by review apps, and the model does not run JavaScript. If the rating renders after page load, the model does not see it.

Audit your top 10 revenue SKUs. View source, count application/ld+json blocks. The ones missing AggregateRating are the ones to ship next.

3. Answer-first PDP copy

The first sentence is what the model uses to summarize the page. “Welcome to our running shoe collection” gives the model nothing. “The Tracer 2 is a 10mm-drop neutral running shoe for flat feet at $145” is a complete recommendation the model can quote.

Test it. Pick your top-revenue PDP. Ask ChatGPT “what is the [product name]?” If the model summarizes it better than your first sentence, your first sentence is the problem.

If you’d rather skip the rewrite, proproductpage.com runs the answer-first rebuild on any product URL. Free check on one product, paid build from $6.

4. Third-party citation density

The model reads what other sites say about yours. A brand mentioned in Modern Retail, Digiday, Adweek, Reuters product coverage, or a founder interview on an operator podcast builds a citation surface the model uses to rank.

The brands ChatGPT names in DTC apparel today were profiled by primary publications between 2022 and 2025. Three to five mentions across that window is enough to anchor a brand inside the recommendation set. Zero mentions is why a brand never appears.

Medium posts, SEO blogspam, and aggregator sites add noise, not signal. One Modern Retail feature beats 50 content-mill mentions.

5. A multi-engine query harness

The four mechanics above are building work. The fifth is measurement. Fire your top 50 to 100 non-branded product-discovery queries against ChatGPT, AI Mode, AI Overviews, and Copilot every week. Log which brands appear in each response. Track delta over time.

Build it in-house with a Node script and the OpenAI / Anthropic / Google APIs, or use Searchable or Profound. Without the harness, the foundation work is faith. With it, you have a number that moves week over week.

Where to start

If you do nothing else from this post, do two things.

Open robots.txt on your store. If GPTBot and OAI-SearchBot are not allowed, fix that today. One line in robots.txt.liquid.

Pick your top-revenue SKU. Ask ChatGPT “what is the [product name]?” If ChatGPT’s summary is more useful than your first sentence, rewrite that sentence this week.

Two changes move enough brands from invisible to occasionally-mentioned that the rest of the stack starts paying off.

If you want this signal four times a week, get the Wire.

Correction policy: if anything in this post is wrong, we’ll fix it publicly with a date-stamped note. Email corrections to support@paidaisearch.com.