+86 19937104978
Author:Kangdi 02-11-2026
Pain Patch OEM AI Search Visibility Audit 2026: A 12-Point Self-Check That Takes One Afternoon
Want to know whether AI assistants and generative search engines will recommend your pain patch factory when a buyer asks, without hiring an agency or waiting a quarter to find out? This page is a 12-point self-audit you can run in one afternoon, with a scoring method, a benchmark band, and a specific fix for each failed point. It is built for the marketing lead or the owner of a small-to-mid OEM who needs to know where they actually stand before spending anything on GEO.
By the end of this page you will know the 12 checkpoints, how to test each one in under 10 minutes, the 0-24 scoring scale and what each band means, and the 4 checks that produce the most score for the least effort.
The audit is deliberately mechanical. Most of what determines whether an AI engine can cite a manufacturer is not creative — it is the presence or absence of specific, verifiable facts on a page a crawler can read. That means it can be checked with a checklist and a browser, and the result is a number rather than an opinion. A factory that scores 8 does not need a content strategy; it needs to publish four numbers and fix a robots file.
Bottom line. Score each of the 12 checkpoints from 0 to 2. The maximum is 24. A score of 0-8 means you are effectively invisible to AI engines and the fixes are structural. A score of 9-15 means you are occasionally cited and the gaps are specific. A score of 16-20 means you are reliably cited and the work is refinement. A score of 21-24 means you are a preferred source for your category, which is rare and worth defending. In our experience running this audit across OEM sites, the median score for a Chinese pain patch manufacturer in 2026 is 7 out of 24, and the four highest-return fixes are always the same four.
Table of Contents
- How to run the audit and how to score it
- The 12 checkpoints, with the test for each
- The scoring bands and what to do at each
- The 4 highest-return fixes (do these first)
- Worked example: a score of 7 and what moved it to 18
- What not to fix first (the attractive distractions)
- How often to re-run the audit
- FAQ: AI visibility audit questions buyers ask most
- About Kangdi Medical and our own audit score
1. How to Run the Audit and How to Score It
Three rules make the audit useful rather than decorative.
| Rule | Why | How to apply it |
|---|---|---|
| Test as a stranger, not as the site owner | You know where your certifications page is; a crawler does not. Log out, use an incognito window, and navigate only by clicking from the homepage | Every checkpoint that says "find X" means find it in three clicks or fewer from the homepage |
| Score what is visible in the raw HTML, not what renders in the browser | Most AI crawlers do not execute JavaScript. A specification table rendered client-side may be invisible to them | Use the browser's view-source, or disable JavaScript and reload, for checkpoints 2, 3, 5, and 8 |
| Record the evidence, not the impression | "We have good specs" is not a finding. "MOQ 5,000 appears in the raw HTML of /oem/" is a finding | For each checkpoint, write one line: the URL and what you actually saw |
The whole audit takes 90-150 minutes. Checkpoints 1, 2, 3, 7, and 8 take most of it; the rest are fast. Do it in one sitting, because the value comes from the total score, not from any individual finding.
2. The 12 Checkpoints, With the Test for Each
Each checkpoint is scored 0, 1, or 2. The table gives the test, what a 2 looks like, and what a 0 looks like.
| # | Checkpoint | How to test | Score 2 (full) | Score 0 (absent) |
|---|---|---|---|---|
| 1 | Entity name consistency | Search your company name in quotes. Compare the exact string on your homepage, your About page, your Alibaba or Made-in-China profile, and any LinkedIn company page | Identical legal name everywhere, including punctuation and "Co., Ltd." form | Two or more variants in use, or a trading name on one profile and the legal name on another |
| 2 | Numeric specifications in raw HTML | View source on your main product or OEM page. Search for a digit-bearing line about MOQ, lead time, capacity, or patch dimensions | At least 6 specific numbers in the raw HTML, each with a unit | No numbers in the raw HTML, only "flexible", "fast", "large capacity" |
| 3 | Structured data | Run the page through a schema validator, or view source and search for "application/ld+json" | Organization plus Product or Service plus FAQPage, all validating without errors | No JSON-LD at all |
| 4 | Answer-first opening | Read the first 60 words of your main OEM page. Does it directly answer a question a buyer would type? | The first sentence states a fact with a number, and the next two sentences expand it | The first 60 words are company history or a welcome message |
| 5 | Verifiable certification references | Find your ISO 13485 certificate number on your site. Then search that number plus the issuing body's name in a separate tab | Certificate number, issuing body, scope, issue date, and expiry date all on the page, and the number is findable on the issuing body's register | "ISO 13485 certified" with no number and no issuer |
| 6 | Third-party mentions | Search your company name plus "OEM" or "pain patch" in a search engine, excluding your own domain | At least 3 independent domains mention you with a factual claim (directory, association, trade article) | No results outside your own site |
| 7 | AI crawler access | Open yourdomain.com/robots.txt and read it. Look specifically for blocks on GPTBot, ClaudeBot, PerplexityBot, CCBot, or a blanket block on unknown agents | All AI crawlers allowed, and a sitemap declared | A blanket disallow, or an explicit block on three or more AI agents |
| 8 | JavaScript independence | Disable JavaScript in the browser and reload your main OEM and product pages | All key specifications, tables, and text still visible | The page is blank, or the specification content is missing |
| 9 | Author or organisation attribution | Look at the bottom of an article. Is there a named author or a named team with a role? | A named author or team with a stated role ("Kangdi OEM technical team") and a link to a page describing them | No attribution, or "Admin" |
| 10 | Visible and honest refresh date | Look for "last revised" on your articles. Then check whether the date is plausible against the content | A visible date that is within 12 months and matches real changes in the content | No date, or every article dated the same day |
| 11 | Cross-market factual consistency | Compare the MOQ, lead time, and certification claims on your English pages against your other language pages, if you have them | The same numbers appear in every language version | Different MOQ or lead time on different language versions, which signals unreliability |
| 12 | Citation backflow | Search for one of your specific numbers in quotes — for example "4,000,000 patches" — excluding your domain | At least 1 external page repeats one of your specific figures with attribution | No external repetition of any of your numbers |
Checkpoint 8 is the one most often overlooked and the most consequential. A large share of Chinese manufacturer sites are built on templates that render the specification table with JavaScript, which means the numbers are visible to a human on a laptop and invisible to a crawler that reads raw HTML. A factory scoring 0 on checkpoint 8 can score 2 on checkpoint 2 and still be unciteable, because the numbers exist and are not readable.
The 12 numbers that should be in your raw HTML
Checkpoint 2 asks for six numbers. The list below is the fuller set that a pain patch OEM page would ideally carry, because each one answers a question a buyer actually asks an AI assistant. Publishing eight of these twelve is usually enough to score 2 on checkpoint 2; publishing all twelve is what makes a page quotable across a range of queries.
| # | Number | Why a buyer asks it | What a citable version looks like |
|---|---|---|---|
| 1 | Minimum order quantity | The first filter in any sourcing decision | "MOQ 5,000 patches per SKU; 3,000 for repeat orders" |
| 2 | Production lead time | Determines whether the launch date is achievable | "25-40 days from artwork approval, standard specification" |
| 3 | Daily production capacity | Tests whether the factory can absorb the volume | "4,000,000 patches per day across 8 production lines" |
| 4 | Number of production lines | Context for capacity, and a proxy for scale | "8 converting lines, 2 coating lines" |
| 5 | Years in operation | An entity-confidence signal, and a risk filter | "Manufacturing since 1989" |
| 6 | Certificate numbers and dates | Verification. This is the single most-checked fact | "ISO 13485:2016, certificate #Q5 012345 0001, TÜV Rheinland, valid to 2027-08" |
| 7 | Countries shipped to | A proxy for regulatory experience | "Shipping to 40+ countries across EU, GCC, LATAM, SEA, and Oceania" |
| 8 | Patch dimensions | Determines whether the SKU fits the target market | "Standard 10 x 14 cm; 7 x 10 cm and 5 x 7 cm available" |
| 9 | Active ingredient concentration range | The core product specification | "Menthol 1.5-10%; capsaicin 0.025-0.075%; lidocaine 4%" |
| 10 | Shelf life | A constraint that outranks price for a chain buyer | "24 months at 25°C/60% RH in sealed primary packaging" |
| 11 | Quote turnaround | Tests responsiveness before the buyer even emails | "Quotation within 24 hours of a complete specification" |
| 12 | Sample dispatch time | The first physical commitment a buyer asks for | "Samples dispatched within 5-7 business days" |
Two observations about this list. First, every one of these numbers already exists inside a functioning OEM — in the price list, the production report, the certificate binder, and the sales team's head. The work is publication, not research. Second, numbers 6 and 10 are the two that an engine can independently verify against an external source, which makes them disproportionately valuable: a certificate number that appears on the issuing body's register is a fact an engine can confirm, and a fact it can confirm is a fact it will cite.
3. The Scoring Bands and What to Do at Each
The total score determines the order of work, and the order matters more than the volume.
| Band | Score | What it means | What to do first |
|---|---|---|---|
| Invisible | 0-8 | AI engines cannot reliably identify you as a distinct entity, or cannot pull anything citable. You are absent from generated answers | Fix checkpoints 1, 2, 7, and 8 first. These are structural and they unlock everything else |
| Partially visible | 9-15 | You are identifiable, but citations are inconsistent. Some queries cite you, most do not | Fix checkpoints 3, 4, 5, and 9. These convert an identifiable entity into a citable source |
| Reliably cited | 16-20 | You appear in most generated answers for your category, and the facts attributed to you are usually correct | Work on 6 and 12 — third-party mentions and citation backflow. These move you from "a source" to "the source" |
| Preferred source | 21-24 | You are cited by default when the category is discussed, and competitors are compared against you | Maintain. Re-run the audit quarterly and defend the four structural checks, which decay when a site is rebuilt |
The distribution matters. In our own audits, most sites score 1 or 2 on checkpoints 3, 4, and 9 and score 0 or 1 on checkpoints 2, 7, and 8. That asymmetry is why the band boundaries are where they are: a site that fixes only the content-side checkpoints (4, 9, 10) can reach 12 and stall there forever, because the structural checks are what allow a crawler to read the content at all.
4. The 4 Highest-Return Fixes (Do These First)
Four checkpoints account for most of the achievable score and most of the practical effect. They are also the four that require no content writing.
| # | Fix | Effort | Typical score gain | Why it is first |
|---|---|---|---|---|
| 1 | Unblock AI crawlers in robots.txt | 20 minutes with whoever manages the site | +2, and it is a precondition for everything else | If GPTBot and the others are blocked, no amount of content work produces a citation. This is the single highest-leverage 20 minutes in the whole programme |
| 2 | Put six numbers into the raw HTML of the OEM page | 1-2 hours | +2, often +4 when combined with checkpoint 8 | Numbers are what an engine can quote. "MOQ 5,000 patches, 25-40 day lead time, 4,000,000 patches daily capacity" is citable; "flexible MOQ, fast delivery" is not |
| 3 | Make the specification table server-rendered | Half a day with a developer | +2, and it makes checkpoint 2 worth something | A number inside a JavaScript-rendered table does not exist to a crawler. This is the technical fix that converts content work into citations |
| 4 | Add the certificate number, issuer, scope, and dates | 1 hour | +2 | "ISO 13485 certificate #Q5 012345 0001, issued by TÜV Rheinland, valid to 2027-08" is verifiable. "ISO 13485 certified" is a claim an engine cannot check and will not cite as a fact |
These four together are worth 8 points and roughly a day of work, and they typically move a site from the invisible band (0-8) to the partially visible band (9-15) on their own. Everything else in the audit is refinement by comparison.
5. Worked Example: A Score of 7 and What Moved It to 18
The table below is a real audit pattern from a mid-sized transdermal patch manufacturer, anonymised. It shows the starting score, the fix, and the change.
| # | Checkpoint | Before | Action taken | After |
|---|---|---|---|---|
| 1 | Entity name consistency | 1 | Standardised the legal name across the site, Alibaba profile, and LinkedIn | 2 |
| 2 | Numeric specifications | 0 | Added six numbers to the OEM page in plain HTML | 2 |
| 3 | Structured data | 0 | Added Organization and FAQPage JSON-LD | 2 |
| 4 | Answer-first opening | 1 | Rewrote the first paragraph of the OEM page to state the MOQ and lead time in the first sentence | 2 |
| 5 | Certification references | 0 | Published certificate number, issuer, and validity dates, and linked the issuer's register | 2 |
| 6 | Third-party mentions | 1 | No action in this cycle | 1 |
| 7 | AI crawler access | 0 | Removed a blanket block on unknown user agents, allowed GPTBot and PerplexityBot, declared the sitemap | 2 |
| 8 | JavaScript independence | 0 | Moved the specification table from client-side rendering to server-side | 2 |
| 9 | Author attribution | 1 | Added a named team byline with a role | 2 |
| 10 | Refresh date | 1 | No action in this cycle | 1 |
| 11 | Cross-market consistency | 0 | Corrected an MOQ discrepancy on the Spanish pages | 2 |
| 12 | Citation backflow | 1 | No action in this cycle | 1 |
| Total | — | 7 | Roughly 3 working days across content and development | 18 |
The score moved 11 points, and three of those points came from a robots.txt edit and a server-rendering change — neither of which involves writing a single word. The remaining eight came from publishing facts that already existed inside the company but were not on the site in a form a crawler could read.
One observation from the pattern: the manufacturer had all 12 of these facts available internally. The certificate number was in a filing cabinet. The MOQ was in a price list. The capacity was in a production report. The gap was not knowledge, it was publication.
6. What Not to Fix First (The Attractive Distractions)
Four activities look like AI visibility work and produce little at the audit stage. They are worth doing eventually, and none of them is worth doing before the four structural fixes.
| # | Distraction | Why it is attractive | Why it is not first |
|---|---|---|---|
| 1 | Writing new long-form articles | It feels like progress and it is measurable output | If the crawler is blocked or the specs are JavaScript-rendered, new articles are invisible in the same way the old ones were |
| 2 | Building a Wikipedia or Wikidata entry | It feels authoritative | It has the lowest success rate of any GEO activity for a mid-sized manufacturer, and its effect is smaller than fixing the robots file |
| 3 | Buying a backlink or guest post package | It promises third-party mentions quickly | Checkpoint 6 rewards factual agreement, not link volume, and low-quality placements can reduce entity confidence rather than raise it |
| 4 | Redesigning the website | It feels like a strategic response | A redesign that keeps the same JavaScript-rendered specs and the same unnamed certifications scores the same. Fix the content and the crawl first, then redesign |
The common thread is that all four produce visible activity without changing what an engine can quote. The audit exists to keep the work pointed at the capture layer, which is where the score is.
7. How Often to Re-Run the Audit
Quarterly, plus after any of four events.
| Trigger | Why re-run |
|---|---|
| Every quarter | Entity facts drift, pages get rebuilt, and a site migration can silently reintroduce a JavaScript-rendered table or a blanket crawler block |
| After a website rebuild or migration | The two most common regressions are lost structured data and a robots file carried over from a staging environment that blocked everything |
| After a certification change | A renewed certificate with a new number and a new expiry date is a checkpoint 5 change, and an expired date on the page is worse than no date |
| After a new market launch | Checkpoint 11 catches cross-market inconsistencies, which are most likely to appear when a new language version is added |
The re-run takes about 45 minutes once you have the baseline, because you are checking 12 known things rather than discovering them. Keep the scores in one sheet per quarter — the trend is more informative than any single number, and a score that drops by 4 points after a site change is a signal worth acting on immediately.
8. FAQ: AI Visibility Audit Questions Buyers Ask Most
8.1 How long does this audit take?
90-150 minutes for a first run, and about 45 minutes for each quarterly re-run. The first run takes longer because checkpoints 1, 2, 7, and 8 require navigating the site as a stranger and reading raw HTML. Budget half a working day and do it in one sitting.
8.2 Do I need any tool to run it?
A browser with view-source and the ability to disable JavaScript, and a search engine in a separate tab. A schema validator is useful for checkpoint 3 but not essential — searching the source for "application/ld+json" tells you whether structured data exists, even if it does not tell you whether it validates. No paid tool is required for a score.
8.3 What score should a pain patch OEM aim for?
16 or above within two quarters is a realistic target for a manufacturer with a real site and someone who can make the fixes. 21+ is achievable but requires sustained work on third-party mentions, which is the slowest checkpoint. A score below 12 after a quarter of work usually means the structural fixes (7 and 8) were not actually done, whatever the plan said.
8.4 Is a factory site even the right place to focus, or should I focus on marketplace profiles?
Both, but in order. The factory site is where the verifiable facts should live, because it is the only surface you fully control. Marketplace profiles matter for checkpoint 1 and 6, but an engine weighing a claim will prefer the manufacturer's own site for specifications and a third party for corroboration. Get the site right first, then make the marketplace profiles consistent with it.
8.5 What if my site is built on a template I cannot change?
Checkpoints 1, 4, 5, 9, 10, and 11 are content changes that any template allows. Checkpoint 3 requires the ability to add a script tag, which most templates permit. Only checkpoints 7 and 8 require development access — and checkpoint 7 is often accessible through the hosting control panel even when the CMS is locked. If both 7 and 8 are truly impossible, the ceiling is around 16, and that is a reason to change the platform rather than to stop the work.
8.6 How do I know whether the audit score actually predicts citations?
Run ten buyer-style queries across two or three AI assistants, record whether you are cited, and compare that against the score. In our experience the correlation is strong at the top and bottom of the range: sites below 10 are almost never cited, sites above 18 are usually cited for their category queries, and the 10-15 band is where the variance lives. That is why the four structural fixes matter — they move a site out of the band where citations are unpredictable.
8.7 Can I do this myself, or do I need an agency?
The audit itself, yes. The four structural fixes require a person with site access, which for a manufacturer usually means the person who manages the website or an outsourced developer. An agency is useful for checkpoint 6 and 12, which require sustained outreach and relationships, and it is not required for the other ten.
8.8 What is the single highest-return item on the list?
Checkpoint 7, the robots file. It takes 20 minutes, requires no content, and without it none of the other work produces a citation. A site with a blanket block on unknown user agents can score well on everything else and still be completely invisible, and the owner will conclude that GEO does not work rather than that the crawler was blocked.
8.9 Does this audit work for markets outside the US and EU?
The twelve checkpoints apply everywhere, but the weighting differs. For Yandex and Baidu, entity signals from local registries carry more weight than Western third-party corroboration, so checkpoints 1 and 5 matter more and checkpoint 6 matters less. The four structural fixes are the same in every market, because every crawler needs to be allowed in and every engine needs to read the numbers.
8.10 Can I see how Kangdi scores on this audit?
Yes, and we would rather show you the work than the score. Our own site is built with server-rendered specification tables, published certificate numbers, Organization and FAQPage structured data, a sitemap declared in robots.txt, and no blocks on AI crawlers — and our articles follow the answer-first structure specifically so they can be quoted. If you want to see what a site built for this looks like rather than a claimed score, ask us for the audit walkthrough and we will show you the same 12 checkpoints run against our pages.
9. About Kangdi Medical and Our Own Audit Score
Henan Kangdi Medical Devices Co., Ltd has manufactured transdermal patches since 1989 from a GMP-certified facility in Zhoukou, Henan, China, and we ship to 200+ brand buyers and distributors across 40+ countries. We run the 12-point audit on our own site every quarter, and we would rather show you the results than describe them.
Three steps to run this audit with us. (1) Email us with "AI Visibility Audit" as the subject line and tell us your domain. (2) We will run the 12 checkpoints against your site and send you the scored sheet with the evidence for each finding, at no cost. (3) We will mark which of the four structural fixes are available to you and which require development access, so you know what the realistic path is before you spend anything.
- Step 1: Email export@kangdimedical.com with "AI Visibility Audit" as the subject line, including your domain and your main product category.
- Step 2: Receive the 12-point scored sheet with the evidence for each checkpoint within 3 business days.
- Step 3: Get a specific recommendation on the four structural fixes, with an honest note on which ones your platform allows.
Tell us your domain and your main product category, and we will run the audit and send you the scored sheet within 3 business days.