Raccoon AI AEO Assessment Report
by Novastacks AI

Start with the good news, because it is the precondition for everything else. The homepage is server-rendered: crawlers receive complete content without executing JavaScript. Answer engines can read the site today. A great many sites in this category fail here, and yours does not.

Page experience is the strongest of the three domains on the measures that matter. The homepage records FCP 1.2 s, LCP 1.4 s and CLS 0.006 — layout stability is effectively perfect, against Botpress at 0.133. Overall performance scores 87 of 100, accessibility 89, and search-readiness 92. The one soft spot is script weight: the page remains busy for 5.4 seconds before it is fully interactive, and visual completeness lands at 2.4 seconds. Neither blocks indexing.

This is the layer that is already right. Every gap that follows sits on top of a foundation that works.

Three of the four pages examined open without a top-level page heading: the flagship product page, the pricing page and the blog index. That heading is the first thing an answer engine reads to decide what a page is about, and on your highest-intent pages there is nothing there to read. Both competitors carry one on all four of their pages.

The section headings that do exist are diluted. On the homepage, four of eight section headings are policy links — privacy, information security, customer terms, partner terms — repeated in the footer of every page. On the product page it is four of six, leaving two real content sections to describe your flagship product. The pricing page runs the opposite failure: 111 section headings, because every plan-comparison row is tagged as a section. A machine reading that page sees a hundred sections of one line each rather than a priced product with tiers.

Social preview images are set on the homepage only. The other three pages share to LinkedIn, Slack and messaging apps with no image — on a site whose buyers are reached through exactly those channels.

Your homepage tells machines that you are an organisation, at a postal address, offering a service — the shape of a local business. Neither competitor's homepage declares a postal address at all. Both instead declare themselves as a software product with a customer rating attached, which is the shape an assistant reuses when it assembles a shortlist of platforms. That difference is not cosmetic: it is the difference between being filed as a company and being filed as a product in your category.

Two of the four pages carry no structured data whatsoever, and one of them is the flagship product page — the single page that explains what you sell, and the page ChatGPT cited first when it did find you.

The FAQ material is the sharpest example of value already built and left unreadable. FAQ blocks sit on all four pages and on a dedicated FAQ page, and ChatGPT cited that page when answering about you. None of it is marked up, so an engine has to infer the question-and-answer structure rather than being handed it. Botpress has not marked its FAQ content up either — only Chatbase has, on one page.

A second, unrelated company operates as Raccoon AI at raccoonai.tech, republished at raccoon-ai.com. It is a general-purpose app-building agent, not a customer service platform, and it is far better known in the United States: a Product Hunt launch, a Show HN thread, Dealroom and Tracxn profiles, YouTube coverage.

The consequence is measurable. On the branded prompt — “What is Raccoon AI and what does its AI customer service platform do?” — Google returned an AI Overview that described the other company in full and cited raccoonai.tech, producthunt.com, agentlocker.ai and aiagentslist.com. Your domain appeared nowhere in the first twenty organic results for your own brand name; twelve of those twenty positions belong to the homonym. The same pattern repeats on every platform we checked: 1,090 YouTube results, 307 Reddit results, 9 Product Hunt listings, one G2 marketplace entry — all of them the other company.

State the limit precisely, because it changes what to do about it. ChatGPT, given the identical prompt, opened by saying there are two different products called Raccoon AI, identified yours correctly from the “customer service” qualifier, and cited six of your pages. This is an engine-specific failure, not a universal one — which means it is an entity-resolution problem, not a lost name.

We ran three unbranded buying questions on each of two engines — the best-platform question, the omnichannel question, and an e-commerce long-tail about automating FAQ and order-status replies — and separately checked the three core category terms in organic search. raccoonai.co appeared in none of them. Six prompt tests, three keyword checks, nine chances, zero appearances. With samples this size the honest reading is a strong consistent signal, not a proven law: every test pointed the same way.

What fills that space is worth naming. ChatGPT's category answers return Intercom, Zendesk, Salesforce, Sierra, Decagon, Ada, Gorgias and Freshworks, citing intercom.com in three of six answers. Google's category results return Freshworks, Kore.ai, Helpshift, Lindy, Nextiva and Kustomer, with Botpress at #17 and #13.

The organic footprint tells the same story from the other end. In the United States the domain holds one ranked search term across the entire first hundred results — “raccoon company”, at position 16, worth an estimated 0.24 visits a month. Not one term in the category vocabulary your product is built for. Chatbase holds 2,965 ranked terms and Botpress 5,283 in the same market.

The site is in English and declares a US English locale alongside Japanese and Traditional Chinese. The content behind that locale is not written for a US reader. The newest blog post is “2026 Smart Customer Service Recommendations: A Guide to Comparing and Choosing from Taiwan's Top 6 AI Customer Service Platforms.” The case studies name Taiwanese brands. The buying guides compare the Taiwanese field.

This is a positioning mismatch rather than a quality problem, and it explains a result that would otherwise look strange: a company with real customers, real case studies and a genuinely useful buying guide, invisible in a market it has declared a locale for. An answer engine serving a US buyer weighs a guide to Taiwan's top six platforms as evidence about Taiwan.

It also explains why the comparison prompt is the one place you do surface. “Raccoon AI vs Botpress vs Chatbase” is a question about your market; your Taiwan-framed guides are about a different one. Both competitors publish head-to-head pages naming the vendors a US buyer is weighing — Botpress against Sierra, Fin and Decagon; Chatbase across fifty-plus alternatives — and both of those pages were cited in the AI Overview that ranked you last.

The most useful finding in this audit is the one that is working. Asked the branded question, ChatGPT ran a live web search, recognised that two companies share the name, chose yours from the “customer service” qualifier, and built its answer out of your own pages — product, homepage, about, contact, FAQ and pricing, six citations in one answer. It then repeated your own positioning back accurately: founded in Taipei, operating across Taiwan, Japan, Singapore, Thailand and South Korea.

That tells us three things the score alone does not. Your pages are readable by machines. Your own copy is good enough to be reused verbatim by an assistant. And the entity confusion is not baked into every engine — it is a resolution failure on one surface, which is a different and more tractable problem than a brand nobody can find.

Second bright spot: your pricing page is your most retrievable asset. It is the page ChatGPT cited in the three-way comparison, it is the deepest page on the site at 1,707 words, and it is one of only two pages carrying structured data. It is doing work the rest of the site is not.