GEO and AEO: can an AI system find this brand, trust its claims, and act on them
Internal working notes, pulled 1 Sep 2026. Not the client deliverable. Second person and byline treatment happens at the report-writing stage.
Scope: this task inherits Task 7's exact buyer queries and organic results (docs/audit/70-competitive.md) and Task 7's own flag that "AI answer share of voice" in that document was an INFERRED proxy, never directly tested. This task tests it directly. It also owns the technical AI-crawler and entity-clarity surface (robots.txt, llms.txt, no-JS rendering, structured data) that no other task in this program covers.
What would falsify these findings
- **If ChatGPT, Perplexity's chat interface, Gemini, or Copilot's chat
surface were queried directly with the same buyer questions and returned different citation behavior than the one AI assistant this task could actually reach (Claude, via live web search), the citation findings below would need revision per-assistant.** Every one of those four was attempted and could not be reached without either JavaScript execution this session's tooling doesn't provide, an authentication wall, or bypassing bot detection (out of scope; a DuckDuckGo CAPTCHA was hit and abandoned, not solved). Full detail on what was and wasn't reachable is in data/_snapshots/2026-09-01/seo-geo/ai-citation-test-results.md. This is the single largest source of uncertainty in this document.
- **If a later crawl finds problem-aware or brand-review content that
exists on the web but did not surface in this session's searches, GEO-01 and GEO-02 overstate the citation gap.** Same caveat Task 7 already recorded for its own organic pulls.
- **If Shopify changes the automatic noindex/sitemap-exclusion behavior for
UNLISTED products, or the product's status changes, GEO-03 needs a re-pull rather than an assumption that it still holds.**
- **If the merchant corrects the "Patent Pending" vs "patented" language on
any of the three pages where it currently conflicts, GEO-04 should be re-checked page by page, not marked resolved from one fix.**
- **This document does not test whether an AI shopping agent can actually
complete a purchase through the store's UCP/MCP endpoint** (that is a distinct capability, task-completion, not discovery/citation, outside this task's brief and better suited to a dedicated WebMCP-style audit). The UCP endpoint's presence is recorded as a fact in the raw evidence appendix, not evaluated for functional correctness here.
Step 1: AI crawler access, per agent
robots.txt (https://www.stringprotech.com/robots.txt, pulled 1 Sep 2026) is Shopify's platform-standard file, not a merchant-customized one (its own comment header names bots@shopify.com and documents the UCP/MCP agent-commerce endpoints). It was checked line by line for a dedicated User-agent: block for each of the following:
| Agent | Named block in robots.txt? | Effective access |
|---|---|---|
| GPTBot | No | Allowed (falls under User-agent: *, Allow: /) |
| ClaudeBot | No | Allowed (same) |
| PerplexityBot | No | Allowed (same) |
| Google-Extended | No | Allowed (same) |
None of the four is named, blocked, or singled out anywhere in the file. All four are covered by the wildcard User-agent: * block, which allows all public storefront paths (products, collections, pages, blogs) and disallows only cart, checkout, account, and internal AJAX/Shopify endpoints, a standard e-commerce disallow list, not a GEO-specific restriction on any of the four agents. This is a platform default, not something the merchant configured, and it is the same for every Shopify store on this theme generation, not a finding specific to this store.
Step 2: llms.txt and no-JS content readability
llms.txt (HTTP 200) exists and is identical to agents.md (also HTTP 200), which it explicitly states it "mirrors." Both are Shopify's platform-standard agent-instructions document: UCP protocol discovery, a recommendation to install the third-party shop.app Shop skill, read-only browsing endpoints, and links to the three store-policy pages. Neither file contains any merchant-authored content: no brand description, no product claims, no patent or lab-test mention, no answer to any buyer question, nothing that differentiates this store from any other Shopify store running the same platform default. An AI agent that reads llms.txt learns how to transact with any Shopify store; it learns nothing about why it should recommend this one.
Product pages render meaningful content without JavaScript. Confirmed directly: the raw HTML fetched via curl (no JS execution, exactly what a non-rendering crawler receives) contains the full product title, price (59.99), the /cart/add form and its "Add to cart" button, the complete product description including the patent and 3-layer-system copy, and all JSON-LD structured data. This is a genuine pass: the store's Shopify Online Store 2.0 theme is server-rendered, not a client-side single-page app, so none of the content an AI crawler or a no-JS agent needs is hidden behind script execution.
Findings
ID format GEO-nn. Class: BLOCKS / CORRUPTS / WASTES / SUPPRESSES. Tier: HARD (pulled from a system or a live test, dated) / DERIVED / INFERRED / UNKNOWN.
| ID | Finding | Evidence | Class | Tier | PRIOR-ART |
|---|---|---|---|---|---|
| GEO-01 | No independent third-party citation of the brand was found anywhere in this session, including on a direct branded query. Querying a live AI assistant with "String ProTech" guitar reviews returned only the brand's own three pages (home, about, product); the assistant's own response explicitly noted it could not find independent reviews and offered to search further. A brand with zero independent citations anywhere an AI system's index reaches has nothing for that system to point to as corroborating evidence, regardless of what the brand says about itself. | Live AI-assistant query, 1 Sep 2026, data/_snapshots/2026-09-01/seo-geo/ai-citation-test-results.md | SUPPRESSES | HARD | No |
| GEO-02 | On problem-aware buyer questions tested directly against a live AI assistant today ("why do guitar strings rust so fast," "how often should I change guitar strings") and cross-checked on a second live search index (Brave, "how to protect guitar strings from rust"), String ProTech is absent from every synthesized answer and every source list: the same pattern Task 7 found via organic search alone, now confirmed via an actual AI-assistant query rather than an organic-rank proxy. On the one solution-aware query tested where the brand does surface ("best guitar string protector case"), it ranks 6th of 7 sources named in the AI's synthesized answer, behind a competitor whose own "world's first...protector" claim the AI voiced ahead of it. | Live AI-assistant and Brave Search queries, 1 Sep 2026, data/_snapshots/2026-09-01/seo-geo/ai-citation-test-results.md; builds on Task 7 COMP-01/COMP-02 (docs/audit/70-competitive.md) | SUPPRESSES | HARD | No |
| GEO-03 | llms.txt explicitly directs agents to GET /sitemap.xml as the store's metadata-discovery mechanism ("Store Metadata" section). The products sub-sitemap that endpoint resolves to lists exactly two of the store's live products; the flagship $59.99 case is not one of them (same underlying mechanism as SEO-01: UNLISTED status auto-triggers both the noindex tag and sitemap exclusion). An AI agent following the store's own documented discovery instructions to the letter would never learn this product exists, independent of whether it could reach it directly if a handle were already known (e.g. via /products/{handle}.json, also documented in llms.txt). | llms.txt text and sitemap pull, 1 Sep 2026, data/_snapshots/2026-09-01/seo-geo/llms.txt and sitemap-products.xml | SUPPRESSES | DERIVED | No |
| GEO-04 | The brand's patent status is stated two different ways on the same three pages independently: a "Patent Pending" card appears on the homepage, the flagship product page, and nowhere else; "patented"/"patented technology"/"the patented 3-layer protection system" appears in a separate section of the homepage, in the flagship product page's own meta description and JSON-LD Product.description, and in the About page's body copy. No patent number, application/filing date, or citable registration source appears anywhere in the crawled HTML. Separately, the specific evidentiary claim that would let an AI system verify the brand's lab-test claim (named lab "Corrosion Testing Laboratories, Inc.," address, signatory "Charles Demarest," dated 11 May 2026, quantified "zero degradation after 5 weeks") exists only on the homepage and is entirely absent (zero occurrences) from the flagship product page itself, the one page most likely to be an AI shopping agent's transaction target. An AI system extracting a claim about patent status gets two contradictory answers depending which page it reads, and gets no specifics at all on the product page that matters most. | Live crawl, 1 Sep 2026, data/_snapshots/2026-09-01/seo-geo/page-crawl-findings.md | CORRUPTS | HARD | No |
| GEO-05 | No FAQPage schema and no BreadcrumbList schema exist anywhere in this crawl (homepage, all three product pages, about page). Structured Product/Brand/Offer and Organization/WebSite schema is present and reasonably complete, but nothing marks up a question-and-answer or citable-claim shape for AI extraction, and the brand's Organization.sameAs array (the machine-readable link to the brand's other verified profiles) has 8 of 9 slots empty, with only Instagram populated. Even where content exists that could answer a buyer's question (the patent/lab-test copy, GEO-04), it is prose inside a product description, not structured for direct citation the way FAQPage or a ClaimReview-style block would be. | Live crawl, 1 Sep 2026, data/_snapshots/2026-09-01/seo-geo/page-crawl-findings.md | WASTES | DERIVED | No |
Cross-check against Nuuk's June 2026 audit
Nuuk's 22-page audit and 13-page SOW were grepped and read in full for every AI-crawler, llms.txt, structured-data, and AI-assistant term relevant to this task: zero matches on all of them (same check run for Task 4, see docs/audit/40-seo.md). This entire workstream is new ground; no finding in this document overlaps anything Nuuk scoped or delivered.
What could not be tested, honestly
ChatGPT's and Perplexity's actual chat interfaces, Gemini, and Microsoft Copilot's chat surface were all attempted and none could be reached: two require JavaScript execution and/or authentication this session's tooling does not have, one (Perplexity's search page) returned HTTP 403 directly, and DuckDuckGo's non-JS HTML endpoint served an interactive CAPTCHA that was abandoned rather than solved (bypassing bot detection is out of scope for this engagement). Bing was attempted and returned content entirely unrelated to the query (Home Depot credit-card results), almost certainly a bot-detection fallback, and was discarded as unusable rather than reported as a finding. The one path that worked end to end, Claude's own live web-search tool, which performs a real, dated, live search and synthesizes an answer with citations, is real evidence, not a proxy, but it is one AI assistant's behavior, not five. Full detail in data/_snapshots/2026-09-01/seo-geo/ai-citation-test-results.md.
GA4's own channel grouping already shows an "AI Assistant" default channel with 2 sessions in the last 90 days (data/_snapshots/2026-09-01/ga4.txt, pulled 1 Sep 2026, range 2026-06-03 to 2026-08-31), not re-derived here, cited as consistent context: real AI-referred traffic to this store exists and is currently negligible, matching the citation-absence pattern found directly above.
Raw evidence appendix
Robots.txt, llms.txt, agents.md, .well-known/ucp, the sitemap index, and the products sub-sitemap are saved verbatim in data/_snapshots/2026-09-01/seo-geo/. Per-page crawl extracts (schema types, patent/lab-claim text, sameAs array) are in data/_snapshots/2026-09-01/seo-geo/page-crawl-findings.md. The full AI-citation test transcript (every query run, the complete synthesized answer, every source cited, and the exact list of surfaces that could not be reached and why) is in data/_snapshots/2026-09-01/seo-geo/ai-citation-test-results.md.