Short answer: we asked ChatGPT, Claude and Perplexity 12 buying questions about retail EAS (electronic article surveillance) and RFID loss prevention systems, 3 times each. On the three shortlist questions, Checkpoint Systems and Sensormatic Solutions were each named in 26 of 27 answers. Across all 108 answers, the assistants cited CONTROLTEK’s own website in 15 and named the company in 9. In 6 answers a CONTROLTEK page was the source and the reader never saw the name.
Key facts from the study
Who ran this study
Revenue Experts is an AI consulting firm with teams in Dallas, Skopje and Toronto. We help established B2B companies find out whether ChatGPT, Claude and Perplexity name them when buyers ask for a shortlist, and whose pages those assistants trust instead.
We work by one rule: a claim you cannot trace is a claim you cannot trust. Every study follows the same published method. We study one buying category at a time and record two things for every AI answer: which companies the assistant names, and which pages it uses as its source. This is our fourth public study. The earlier ones covered retail point-of-sale software (NRS), AI governance software (OAIS) and field service software (ServiceTitan). This one covers retail loss prevention.
Why retail EAS and RFID?
EAS and RFID systems are the tags, labels, detection pedestals at the store door and detachers that retailers use to cut theft. We wrote the questions for one buyer: a VP or Director of Loss Prevention at a US retail chain of roughly 50 to 1,000 stores.
In the answers we collected, two vendors dominated and a group of challengers followed, which makes it a fair test. If an established challenger can be read by the assistants and still go unnamed, smaller companies in any category should check their own position.
One honest note before the findings. Chain-level EAS buying runs largely through direct sales, integrators and industry bodies such as LPRC and NRF Protect. AI answers may weigh less in this purchase than in categories bought by online comparison. We wrote that down before the run.
What did the assistants name on the shortlist questions?
Three of the twelve questions ask directly for a shortlist:
- “Best EAS systems for a mid-size chain across a few hundred stores”
- “Tagging and detection vendors for an apparel chain that can run a source tagging program with suppliers”
- “Which providers can standardize one system across a few hundred stores in different formats”
Across the 27 answers to those three questions:
CONTROLTEK’s own blog post, “Understanding retail shrink: the importance of an effective EAS system”, was cited as a source in 2 of those 27 answers, both on Perplexity. The company’s name did not appear in either.
The pages that carried the shortlist were direct answer pages written in the buyer’s terms. Sensormatic’s EAS overview page alone was cited in 12 of the 27 shortlist answers. Source-tagging pages from Sensormatic, Checkpoint and ALL-TAG carried the source-tagging question. Apparel solution pages carried the apparel question.
Where did CONTROLTEK appear?
Across all 12 questions, CONTROLTEK was named in 9 of 108 answers and its own site was cited in 15 (20 citations in total). That puts it 5th of 13 ranked brands by name and 4th of 13 by citation. Every answer that named it also cited its site.
The assistants trusted CONTROLTEK’s pages. They did not always hand the reader its name.
Three details explain most of the result:
- One question carried it. On “which EAS and RFID vendors in the US offer nationwide service technicians and fast tag resupply”, Claude and Perplexity named CONTROLTEK in 3 of 3 runs each. That one question holds 6 of its 9 named answers. Its solutions page was cited 7 times there, because the page answered that exact question.
- Its own website did the work. Every answer that named CONTROLTEK also cited its own site. Competitors also reached answers through trade media, resellers and integrators. In the pages we could check, CONTROLTEK’s appearances rested on its own website.
- Each assistant treated it differently. By assistant, of 36 answers each: Claude named CONTROLTEK 6 times and cited it 10. Perplexity named it 3 times and cited it 5. ChatGPT treated it differently again, which is why we report each assistant separately.
The market: named versus cited
For each brand we counted the answers that named it and the answers that cited a page on its own website. Of 108 answers:
The most cited websites in the whole study were sensormatic.com (130 citations), securitytagstore.com, a reseller (66), checkpointsystems.com (43) and alibaba.com (43). The trade site losspreventionmedia.com was cited 24 times. The study held 1,401 citations in total.
The assistants did not agree with each other
Same questions, same day, same brands, and three different pictures of the market.
The three assistants treated the same vendor differently. Claude and Perplexity named CONTROLTEK every time on the service question; ChatGPT did not. We also measured how closely each assistant’s “named” ranking of brands matched its “cited” ranking. On ChatGPT the two lists agreed closely (0.924). On Claude they agreed less (0.835). On Perplexity they agreed least (0.504): the brands it named and the brands whose pages it used were often different.
A single AI visibility score averaged across the three would hide all of this.
What each competitor’s result teaches
Sensormatic Solutions: one page that answers the shortlist question. Named in 45 answers, its own site cited in 62, and sensormatic.com was the most cited site in the study with 130 citations. Its EAS overview page alone was cited in 12 of the 27 shortlist answers. What it means for you: in this study, the vendor with the clearest page for the buyer’s exact question tended to become the source for that question.
Checkpoint Systems: a name that travels beyond its own pages. Named in 40 answers, its own site cited in 29, and named in 26 of 27 shortlist answers. Its name reached more answers (40) than its own pages did (29). What it means for you: a place on the shortlist likely depends on what other sites say about you, as well as on your own pages.
Nedap and Avery Dennison: names and pages in step. Nedap was named in 20 answers and cited in 17; Avery Dennison 13 and 13. What it means for you: when your pages and your name show up together, the assistant has both the fact and the vendor to credit.
ALL-TAG: a specific page beats a small footprint. Named in 5 answers and cited in 9, ALL-TAG’s source-tagging pages helped carry the source-tagging question next to the two most named brands. What it means for you: one page on a narrow buying question can put a smaller company into the answer.
Zebra, Alpha High Theft Solutions, Honeywell and NXP: known, but someone else tells the story. Each was named (6, 5, 4 and 4 answers) and none of their own pages was cited. What it means for you: if the assistant knows your name but uses other people’s pages, you do not control how it describes you.
Retail Security Group: heavily cited, rarely named. Its own pages were cited in 28 answers, but it was named on only one question, so our ranking rule left it out. It shows CONTROLTEK’s pattern at a larger scale. What it means for you: high citation counts can hide the fact that buyers never see your name.
securitytagstore.com and losspreventionmedia.com: the third-party route. A reseller’s site drew 66 citations and a trade publication 24. Competitors reached answers through resellers, integrators and trade media. What it means for you: the sites your buyers’ assistants read are not only yours.
What this means for any B2B company in this position
We saw versions of this position in our NRS and ServiceTitan studies too: pages used as sources without the company’s name. Here, the assistants know the company well enough to use its pages, and on one question they name it every time. When the buyer asks the question that matters most, “who should be on my list”, the answer goes to the brands whose pages answer that question directly and whose names appear on other sites.
Three lessons carry over to any category:
- A citation without your name does little for you. If your pages explain the topic but never say plainly what you sell and for whom, the assistant can use the explanation and leave the vendor out.
- Strength on one question does not spread to the others. CONTROLTEK won the service question because one page answered it. Nothing answered the shortlist questions in the same way.
- Your own website is one route of several. Competitors appeared through trade media, resellers and integrators. CONTROLTEK appeared only through its own pages.
Six steps you can take this week
These are our recommendations, built on what the study shows. They are well-founded bets to test in your own category. None of them guarantees a result.
- Check your security wall. CONTROLTEK’s homepage, solutions page, contact page and robots.txt returned a security challenge to every automated request from our network. Confirm whether your own filter blocks non-US traffic, all automated visitors, or both, and whether it lets legitimate AI crawlers and assistants through. Publish a robots.txt that anyone can read. First step: ask your IT or web team to open your robots.txt from outside your office network. (Its effect on the assistants is untested.)
- Write one direct answer page per shortlist question. Use the buyer’s own words in the title, answer in the first screen, and name your offer in the page. First step: write down the three shortlist questions your buyers ask and the page on your site that answers each. A blank line is your first page to build.
- Turn citations into names. Next to every explanation on your blog and product pages, say plainly what you offer and for whom. First step: rewrite the first paragraph of your three most-read educational posts so each one names your company and your offer.
- Earn mentions on the sites the assistants read. Trade media, resellers, integrators and directories all fed these answers. First step: list the five trade or partner sites in your category and check whether any of them mentions you today.
- Check each assistant separately. Claude and Perplexity named CONTROLTEK every time on the service question; ChatGPT did not. First step: ask one buying question in ChatGPT, Claude and Perplexity, and mark for each whether you were named and whether your own page was cited.
- Measure every change with a controlled test. Change one page, keep a similar page unchanged, and rerun the same questions before and after. First step: pick the page you will change first and the page you will leave alone, and save today’s answers as your baseline.
Mistakes to avoid while you do this: counting mentions and stopping there, judging all three assistants by one, and changing ten pages at once so you cannot tell which change worked.
Is this you?
If two of these sound familiar, your company may sit where CONTROLTEK sits:
- Your blog explains your category well, but the posts rarely say what you sell.
- Your sales team wins on service or coverage, yet no page on your site answers “who should be on my shortlist”.
- Your competitors appear in trade media and on reseller pages, and you mostly appear on your own site.
- Nobody on your team has asked the same buying question in ChatGPT, Claude and Perplexity and compared the sources.
How the study was run
- Buyer: a VP or Director of Loss Prevention at a US retail chain of roughly 50 to 1,000 stores.
- Questions: 12, brand-neutral, written the way this buyer would ask them. The questions, the brand list and the assistant settings were frozen and pushed to GitHub before the first query.
- Assistants: ChatGPT, Claude and Perplexity through their APIs with web search required. This measures the assistants when they search; the consumer apps may answer differently.
- Runs: 3 per question per assistant, each in a fresh session. 108 answers, all with confirmed web search, run on 26 September 2026.
- Counting: a citation counted for a brand only when the page sat on the brand’s own website. Every other page was logged as third-party. 85 unclear name matches on common words such as Checkpoint, Alpha and Century were ruled by hand by Elizabeta Kuzevska.
- Page check and site access check: run on 26 and 27 September 2026.
The prediction we wrote before the run
Before the first question we recorded a prediction for how closely the “named” ranking and the “cited” ranking of brands would agree: 0.70, with a range of 0.55 to 0.85. The result was 0.764, with an interval of 0.570 to 0.919. By the rule we set in advance, that is inconclusive. The top three brands were the same on both lists. The counts in this article do not depend on that statistic.
What we got wrong or could not check
- Our citation link-resolution step stopped before it finished, so we did not use it.
- The page check had to be rerun with time limits.
- The site access check was added after the run, so it was not part of the plan we froze in advance. Since this study, it is a standard step in every study we run.
- 22 of the 169 cited pages could not be checked, including all 7 CONTROLTEK pages.
- Whether each cited page supports the sentence beside it was not verified in this run.
- Retail Security Group’s pages were cited in 28 answers, but our ranking rule leaves it out because it was named on only one question. The fix to the method is still open.
On site access: the security challenge CONTROLTEK’s pages returned to our network on 26 and 27 September 2026 was a Cloudflare challenge (HTTP 403). The cause is not determined and it may be geographic. The assistants still cited the site 20 times, so this does not show that they were blocked. When we checked again on 9 October 2026, the same pages and robots.txt loaded normally from our network.
What the study did not measure
A brand named in an answer does not prove a recommendation. A cited page does not prove the page supports the sentence beside it. An absence does not prove a lost customer. This is one category, one date, 12 questions and three assistants used through their APIs. The study records who was named and whose pages were cited. It does not show what any buyer did.
Frequently asked questions
Which retail EAS vendors do AI assistants name most often?
In this study, Checkpoint Systems and Sensormatic Solutions were each named in 26 of 27 answers to three shortlist questions, followed by Nedap (20) and Avery Dennison (11). ChatGPT, Claude and Perplexity were each asked every question three times on 26 September 2026.
How often did AI assistants name CONTROLTEK?
Across all 12 questions, CONTROLTEK was named in 9 of 108 answers, and its own site was cited in 15. Claude and Perplexity named it in every run of the question about nationwide service technicians and fast tag resupply.
Is being cited by an AI assistant the same as being named?
No. In 6 answers a CONTROLTEK page was a source and the company was not named. Retail Security Group’s pages were cited in 28 answers, but it was named on only one question.
Which pages do AI assistants cite for EAS shortlist questions?
Direct answer pages written in the buyer’s terms. Sensormatic’s EAS overview page alone was cited in 12 of the 27 shortlist answers. Source-tagging pages from Sensormatic, Checkpoint and ALL-TAG carried the source-tagging question.
Do ChatGPT, Claude and Perplexity treat the same vendor the same way?
Not in this study. Claude named CONTROLTEK in 6 answers and cited it in 10; Perplexity named it in 3 and cited it in 5. How closely the named and cited rankings agreed ranged from 0.504 on Perplexity to 0.924 on ChatGPT.
Could AI assistants reach CONTROLTEK’s website?
They cited it 20 times during the run, so the challenge does not show they were blocked. When we requested the homepage, solutions page, contact page and robots.txt from our network on 26 and 27 September 2026, each returned a security challenge. The cause is not determined. When we checked again on 9 October 2026, the same pages and robots.txt loaded normally from our network.
How many AI answers did the study analyse?
108: 12 questions, 3 assistants, 3 runs each, all with confirmed web search.
Want this study for your own category?
This study used 12 questions. Our audit runs the same method on your category and your competitors. We write your buyers’ questions with you and freeze them before the run, ask them in ChatGPT, Claude and Perplexity, separate being named from being cited, open the pages the assistants cite in your place, check whether AI assistants can reach your site, and hand you the fixes in order. The audit costs $1,495 and takes 2 to 5 days. To talk it through, book a 20-minute fit call with John Bush, Co-Founder of Revenue Experts: https://meetings.hubspot.com/john2956
The easy first step is free: a 12-prompt sample AI visibility report on your category. Message Elizabeta on LinkedIn, https://www.linkedin.com/in/elizabeta-kuzevska/ . If you are not connected, send a connection request with a note asking for the sample report. We reply within three working days. You can also ask through the form on our AEO page: https://revenueexperts.ai/aeo-service/ .
Evidence: the study’s public evidence package: https://drive.google.com/drive/folders/1JW71l1XYM9lL1LdQkWDqN-Q-VAHDjjZN
Method: RE-METH-AIV-003, Case Study Methodology, Public Edition 1.0: https://revenueexperts.ai/wp-content/uploads/2026/10/RE-METH-AIV-003-Case-Study-Methodology-Public-Edition-1.0.pdf