The AI shopping visibility audit: 10 things ecommerce brands should check before the next sales season in 2026

Key takeaways

• AI-referred traffic to US retail sites is forecast to grow 130% YoY this holiday season, and it converts noticeably better than non-AI traffic, so invisibility in AI answers has a real revenue cost.
• Product pages and landing pages make up close to 40% of all ChatGPT citations, and around 70% of citations on commercial/shopping prompts specifically, per Promptwatch Data, so your product page quality matters more than your blog.
• Blocking the wrong crawler (OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot) can erase you from AI shopping answers entirely, and most brands have never checked their robots.txt for this.
• Google now ties AI Mode/AI Overview visibility to specific Merchant Center attributes (related product, product highlight, Q&A, popularity rank) that most feeds don't populate.
• Reddit's share of ChatGPT citations collapsed from roughly 4% to 0.5% in a single day in August 2026, a reminder that AI citation patterns shift fast and need regular rechecking, not a one-time audit.

Why this audit can't wait until Black Friday week

I'll be blunt: if you're reading this in October and planning to "deal with AI search after the holiday rush," you're already behind. Adobe's data (aggregated by CartFlows) puts full-season AI-referred traffic growth at 130% year over year for 2026, following a 693% jump in 2025. That's not a rounding error you fix with a quick patch in November. And the traffic isn't just bigger, it's better: AI-referred visits converted 38% better than other traffic on Black Friday 2025, with the widest gap on Thanksgiving Day at 54% better conversion.

The catch is that most ecommerce teams still treat "AI visibility" like a vague brand-awareness metric instead of a channel with its own technical requirements. It isn't. ChatGPT, Google AI Mode, and Perplexity decide which products to recommend based on crawlable product data, feed attributes, and citation-worthy content, each with its own rules. An audit that only checks "are we mentioned" misses the mechanics that actually determine whether you get mentioned at all.

Below are the 10 things worth checking before the next sales season, roughly in the order I'd tackle them.

1. Can AI crawlers actually reach your product pages?

Start here because it's the most common and most embarrassing failure. OAI-SearchBot and ChatGPT-User are the crawlers that power live product discovery in ChatGPT; if either is blocked in robots.txt, you're invisible in ChatGPT shopping results, full stop. The same logic applies to PerplexityBot/Perplexity-User and ClaudeBot/Claude-SearchBot.

A sane ecommerce robots.txt allows those search/citation bots while disallowing crawl of /cart/, /checkout/, /account/, and faceted-filter URLs like ?sort= or ?filter= for everyone else. Note the distinction within each vendor's bot family: GPTBot trains models, OAI-SearchBot powers search citations, ChatGPT-User fetches pages in real time when a user asks a direct question. You can block the training-only bot and still allow the ones that drive citations, if that's a stance you want to take. Bytespider is worth blocking outright given its documented history of ignoring robots.txt.

Pull your server logs or a crawler-monitoring tool and confirm these bots are actually hitting your product pages, not just that robots.txt technically allows them. A misconfigured CDN rule or a bot-blocking security plugin can quietly override robots.txt.

2. Does your product feed meet 2026's AI Mode requirements?

Google published, in May 2026, eight specific Merchant Center attributes tied directly to AI Mode and AI Overview representation: product highlight, product detail, variant option, item group title, related product, question and answer, document link, and popularity rank. Most feeds built for traditional Shopping ads don't populate half of these.

The "related product" attribute is worth special attention. It's repeatable up to 30 times per product and feeds the "what else do I need" and "are there alternatives" style queries AI Mode handles constantly. If you're not populating it, you're ceding that entire query type to competitors who do.

Also check image sizing now, not later. Google's minimum product image requirement jumps to 500x500 pixels, enforced starting January 31, 2027, which sounds far off until you remember catalog refreshes take months to roll out across a full SKU range.

3. Is your catalog wired into ChatGPT's commerce layer?

OpenAI has pulled back from a standalone Instant Checkout flow and is now prioritizing shopping discovery, with purchases completing on your own site. If you sell on Shopify or Etsy, your catalog is already plugged in through the Agentic Commerce Protocol, no extra application needed. Everyone else has to apply and sits on a waitlist, with a self-serve merchant platform promised but not yet universal.

A product feed isn't strictly required if ChatGPT already crawls your site, but it gives you control over pricing and availability accuracy that crawling alone won't guarantee. Given that shopping features are currently US-only in ChatGPT and the feature set is still shifting, check your merchant status now rather than assuming you're covered.

4. Are your product pages actually citation-worthy?

Per Promptwatch's citation-type data, product pages and landing pages together account for roughly 39% of all ChatGPT citations, and on commercial/shopping-intent prompts specifically that combined share climbs toward 70% once you add listicles and comparison pages. Reviews, by contrast, sit at a thin 1.7% of citations overall, though they jump to 5.6%, three times their average, once a prompt names your brand directly.

The practical read: your product detail pages need clear specs, accurate pricing, and enough descriptive substance to be quoted directly. Thin pages that rely on JavaScript-rendered pricing or vague marketing copy give the AI nothing to cite. If you're running review aggregation, make sure it shows up once customers ask about you by name, since that's exactly when reviews start pulling weight.

5. Have you run your actual shopping prompts, not generic brand queries?

Most AI visibility checks start and end with "does ChatGPT know my brand." Wrong question for retail. Run the specific comparison-shopping prompts your buyers type: "best noise-cancelling headphones under $300," "[category] alternatives to [competitor]," "what should I get my dad for the holidays under $100." Log which brands get named, which get shopping cards, and which get neither.

Promptwatch's shopping-usage data shows ChatGPT's product-card feature doesn't appear at a constant rate, it moved in visible steps during 2026, roughly doubling overnight in late May before falling back weeks later. That means a single snapshot audit tells you almost nothing. You need to run the same prompt set repeatedly through the season to catch when OpenAI opens or closes a visibility window for your category.

Tools like Promptwatch track this kind of prompt-level shopping behavior along with citation trends, so you're not manually re-running queries every week.

6. Who's filling the gap when you're not cited?

An audit that only measures your own visibility misses half the picture. Pull up the same prompts and look at who the AI recommends instead. In the automotive vertical, for instance, seven of the top eight cited domains on ChatGPT are marketplaces and research sites, not manufacturer sites, and even a single local dealer page cracked the top 31 at a 0.28% citation share. The lesson translates directly to retail: marketplaces, review aggregators, and niche category pages often out-cite brand sites, even small or local ones.

If a competitor or a third-party marketplace is consistently filling your product's slot, that's an actionable gap, not just a monitoring data point. It tells you whether your problem is crawlability, content depth, or simply that AI trusts a third-party source more than your own site for that query.

7. Are you tracking the right domain authority band?

Here's a counterintuitive one. Promptwatch's domain-rank data for August 2026 shows ChatGPT citations cluster in the mid-tier, DR 46-75 domains took roughly 46% of citations combined, while the very top tier (DR 91-100) shrank from about 7% to around 3% over the month. Meanwhile DR 0-30 sites still held about 14% of citations.

If your brand site sits at a modest domain rating, that's not automatically a disqualifier the way it would be for classic SEO link-building. Don't assume you need Wikipedia-tier authority to get cited; you need relevant, accurate, well-structured product content more than raw domain strength.

8. Is your off-site presence pulling its weight?

Reddit's share of ChatGPT citations collapsed from around 4% to 0.5% in a single day, August 14, 2026, a sharp reminder that relying on forum buzz for AI visibility is fragile. Social platforms overall now account for less than half a percent of ChatGPT citations, though Google AI Overviews still cites social content at a much higher rate, around 12.2%.

Check where your earned mentions actually live: press coverage, comparison roundups, third-party review sites, category listicles. Since listicles and comparison pages are among the fastest-growing cited formats, getting your product correctly featured in third-party buying guides is worth real effort heading into the season, more so than chasing Reddit threads that may not move the needle for ChatGPT specifically.

9. Does your schema markup qualify you for the features that matter?

Schema.org Product markup gets you into the traditional Merchant listing rich result, but it does not automatically qualify you for Google's stricter "Universal Cart" buy-button eligibility, which launched in the US, Canada, and Australia. Treat these as two separate gates and audit for both.

Also worth a quick check: Google's structured_title and structured_description attributes now carry a digital_source_type sub-attribute distinguishing AI-generated from human-written copy. If you're using AI tools to generate product descriptions at scale, make sure that metadata is set correctly and not accidentally stripped, since the same rule applies to AI-generated product images, which need IPTC DigitalSourceType metadata preserved.

10. Can you actually attribute AI-driven sales?

This is the audit item teams skip because it's annoying, not because it's unimportant. A July 2026 survey found 64% of AI shoppers still open a separate tab to complete a purchase on a website rather than clicking through the AI tool's own link, which quietly breaks a lot of last-click attribution models. If your analytics setup can't distinguish an AI-influenced visit from a direct one, you'll systematically undercount this channel's impact and under-invest in fixing items 1 through 9.

At minimum, tag campaigns and landing pages so you can spot referral patterns from AI platforms in your analytics, and cross-check against any visitor-level AI traffic reporting your visibility platform offers.

Putting the audit on a schedule, not a shelf

A table to keep the priorities straight as you work through this before the season ramps up:

The common thread across all ten items is that AI visibility for retail isn't a one-time fix. Citation shares shift by percentage points in a single day, as the Reddit collapse showed. Shopping card display rates double overnight and fall back weeks later. Running this audit once in October and calling it done means you're flying blind by Black Friday.

If you want a platform that handles the repeated prompt testing, citation tracking, and crawler-log monitoring across the season rather than doing it by hand every week, Promptwatch covers ChatGPT, Gemini, Perplexity, AI Overviews, and AI Mode with shopping-specific tracking and crawler analytics built in. For teams that want to compare more options across the category, the GEO software directory at bestgeosoftware.com and the ecommerce-focused listings on surferstack.com are both worth a look before you commit to a tool for the season.