Boutique Cited by AI 71 Times, June to July 2026

In 30 days, AI answer engines cited my mom's small boutique 71 times. Microsoft Copilot and its partner surfaces, pulling from her pages to answer real questions. Across those queries the shop held about a third of the cited sources, and 73% on the top one. Ad spend behind it: zero. The source is the Microsoft Clarity AI Visibility export, June 18 to July 18, 2026, and I keep the originals.

This is the exact setup, in the order I ran it.

Update, July 20, 2026: the next 30-day window (June 21 to July 20) came in at 80 citations. This case study keeps its original 71 because that is the window it measured; I publish each new number as it lands.

Correction, July 28, 2026. Three fixes after a full audit against the primary exports. One: this article originally named "ChatGPT, Perplexity, Google's AI answers" as the citing engines. Clarity measures Microsoft Copilot and its partner surfaces and does not see those three, so the opening now names the surface the instrument actually covers. Two: "roughly 90%" of citations from two guide pages was measured at 83% at the time; it now reads more than 80%. Three: the 3,033% clicks figure came from Google Merchant Center's Online store report rather than Google Search Console, and it moves as the comparison window fills in. It is corrected below. The full correction list lives at llms.txt. Nothing here was overstated in the other direction: 71 and 80 were both conservative against a later verified 85 for the same window.

The store this happened to

Krafty Bandit is my mom's boutique. Real Shopify store, real products, real customers, and a real budget (almost none). In mid-2026 I ran the kind of technical playbook that usually lives behind enterprise retainers on her shop. Everything below happened to a store you can walk into, not a demo.

Step 1: fix the entity first

AI can only cite a business it can resolve. If your name, address, and phone differ across your site, Google Business Profile, and the directories, the machine's conclusion is "I'm not sure this business exists." I made one byte-identical name-address-phone block, put it everywhere, added LocalBusiness schema carrying it, and fixed a directory listing that had her filed as a "Sewing Store." Boring work. It gates everything else.

Step 2: make each page answer one thing, cleanly

Two clean, single-purpose guide pages earned more than 80% of the shop's AI citations. Not the homepage. Not the product pages. Guides that answer one real question end to end, written in short passages: 50 to 150 words, one idea each, under a heading that says what the passage answers, in plain subject-verb-object sentences. AI answers are assembled from passages. Write passages worth assembling. (This article is written that way on purpose.)

Checked 28 July 2026, and this is the part I could not prove until this week: those two pages have no backlinks. Ahrefs shows exactly one page on the entire domain with external links pointing at it, and that page is the homepage, which is not one of the two. The shop's whole dofollow profile is six referring domains: my own site, one small shop that links to hers, two directories flagged as spam, one local sewing directory, and one business-listing aggregator. Not a single editorial link, and a Domain Rating of 4.9. Whatever earned those citations, authority was not it.

Step 3: the schema stack

Structured data is how you say the quiet part in a language machines trust: LocalBusiness, Product with per-variant Offers, BreadcrumbList, and FAQPage. I rebuilt her 8-question FAQ into 34 questions with matching JSON-LD. One rule I never break: the visible text and the schema have to match exactly, and resale goods never get called "handmade." Schema that flatters you is schema that eventually burns you.

Step 4: open the doors for AI crawlers

Your robots.txt may be blocking the exact bots you want citing you. I explicitly allowed GPTBot, PerplexityBot, ClaudeBot, and Google-Extended, then went further down the agentic-commerce path: agents.md, merchant program enrollments, AI shopping channels. Being cite-able and being buy-able inside an AI answer are different jobs. Correction, July 28, 2026: that agent list is incomplete, and I have fixed it across the rest of the site. OpenAI also runs OAI-SearchBot for appearing in ChatGPT's search results and ChatGPT-User for the live fetch made while answering. Anthropic runs Claude-User and Claude-SearchBot alongside ClaudeBot. And Google-Extended governs Gemini apps and Vertex AI grounding rather than Search, so it is not the AI Overviews lever I once treated it as.

Step 5: catalog hygiene at scale

This is the unglamorous engine. Of 2,861 product images across 1,082 active products, about 990 had no alt text at all. After one weekend, exactly one blank remained. 2,810 junk image filenames got cleaned. 52 wrong vendor labels got corrected. I did it in a weekend because I built a team of Claude-based AI workflows to do the lifting, with a changelog for every change and a human (me) deciding what shipped. That build is its own story, coming as the first video on the channel.

Step 6: measure it or it didn't happen

Three instruments, all free: Microsoft Clarity's AI-citations dashboard (the flagship data above), a GA4 custom channel group that catches AI referrals before they hide inside Organic, and a monthly probe. The probe is simple: the same 10 buyer-style questions asked across 5 engines, logged in a fixed table so the trend is visible. If a consultant can't show you instruments like these, they are guessing.

What the citations did not do

Those 71 citations sent almost no click traffic. About 0% of sessions. AI engines cited the shop on how-to questions; they did not send visitors, and they are not yet recommending it on the "where do I buy" question. Google Merchant Center did go from 0 of 1,795 products approved to the whole catalog live on June 16 to 17, and organic clicks in its Online store report rose about 1,006% against a base of 16 clicks. The base was tiny, so the absolute numbers are small. Real trajectory, small base. Anyone who shows you the percentage without the base is selling something. Sales grew too, and I don't credit this work for that, because attribution is messy. Being the source AI trusts is the position I want her shop in as search shifts, and the traffic gap is the next problem I'm working on in the open.

Do this to your store

Fix your entity. Pick the three questions your customers actually ask and answer each one on its own clean page. Add schema that matches your visible text. Unblock the AI crawlers. Fix your alt text and filenames. Install the free instruments. Then measure for 30 days. If you want to know where you stand before you start, run the 10-minute check to see if AI can read your store; the thinking behind every step is in the plain-English GEO guide. That's the whole starting playbook, and every video on the channel shows one of these steps done to a real store.

Questions readers ask me about this case study

Which AI engines actually cited the shop?

Microsoft Copilot and its partner surfaces, and that is the limit of what my instrument can see. Microsoft Clarity's AI Visibility report covers that family and does not measure ChatGPT, Perplexity, or Google's AI Overviews. So when I say 71 or 80 or 85 citations, I mean within Copilot and its partner surfaces. Whether the same pages get pulled by the other engines is an open question on this store.

Can a small shop really compete with big retailers in AI answers?

On the right questions, yes, and this shop is the example. It never beat anyone on "where to buy." What it won were care and comparison questions where a specific, useful guide page existed and the big retailers had nothing equivalent. A single genuinely useful page can out-earn an entire catalog, because engines are assembling an answer rather than ranking a store.

How much did this cost in ad spend?

Zero on the results claimed here. The citations, the Merchant Center approval, and the organic clicks all happened with no ads behind them. The shop does run a small paid social budget separately, which is exactly why I keep sales out of the causal claim: I can prove the organic wins and I cannot cleanly separate the revenue.