GrackerAI reported that 73% of the 100 cybersecurity vendors in its benchmark received no ChatGPT citations when its researchers tested vendor recommendations within the companies’ own product categories. The result does not mean that 73% were absent from every AI assistant, or that they were never mentioned in any AI-generated answer.
The GrackerAI cybersecurity benchmark, published in February 2026 and promoted in an April 29 announcement, examined 100 vendors and 250 standardised buyer-intent prompts across six AI platforms. The research ran from September 2025 through January 2026. Its headline is a specific finding about ChatGPT recommendations, not a measured rate of invisibility across the wider AI market.
Editorial correction: An earlier version of this article described the 73% result as limited or missing visibility across AI-assisted research generally and suggested possible causes without evidence that the study had tested them. This update clarifies the source, platform, sample and measurement limits. It also removes an unsupported connection to Google’s March 2026 core update.
The denominator behind GrackerAI’s 73% finding
The vendor is the unit of the headline statistic. In the test, 73% of sampled companies reportedly did not receive a citation from ChatGPT in the relevant category-recommendation questions. That is not a claim that 73% of prompts returned no vendors, that 73% of all cybersecurity companies are unrecognised, or that the same companies were absent from the other platforms.
| Study element | What the published report specifies |
|---|---|
| Source and status | GrackerAI’s own 2026 industry benchmark, not an official AI-platform assessment. |
| Research dates | September 2025 to January 2026; report published February 2026. |
| Vendor sample | 100 cybersecurity vendors across 10 product categories, from larger companies to smaller providers. |
| Platforms in the broader study | ChatGPT, Perplexity, Claude, Gemini, Microsoft Copilot and Google AI Overviews. |
| Prompt set | 250 standardised buyer-intent prompts: 80 awareness, 100 evaluation and 70 decision questions. |
| Headline 73% | Vendors with zero reported ChatGPT citations for recommendations in their own category, not an average across six platforms. |
The report includes categories such as endpoint detection and response, SIEM, identity and access management, and managed detection and response. Its prompt examples range from general category recommendations to product comparisons and requests shaped by company size or compliance requirements. These are constructed research prompts, not a record of what every cybersecurity buyer asked in a live purchase journey.
A missing citation is not necessarily a missing brand mention
The word citation deserves particular care. An AI answer might name a supplier without linking to its website, recommend it alongside competitors, or cite an independent review that discusses the supplier. Those events are related but not identical.
GrackerAI’s methodology describes citation frequency partly in terms of how often a brand appears. It does not publish a complete response-level dataset or a sufficiently detailed matching rule to let an independent reader reconcile all brand mentions, recommendations and clickable source links. It would therefore be an overstatement to translate its zero-citation finding into proof that a company was never recognised or recommended.
A vendor can also appear in one prompt and be absent in another because the requested category, buyer constraints or available source material differ. The article’s original assertion that smaller vendors were necessarily overlooked because of weaker entity signals was not established by this benchmark. Nor does the study show that all vendors without ChatGPT citations lost buyer enquiries or revenue.
Broader changes in referral patterns are a separate subject. MOCOBIN’s AI visibility and referral traffic analysis deals with that wider measurement question. It should not be used to reinterpret this vendor-level result as a traffic study.
What the commercial research can and cannot explain
The report is useful for showing how a particular benchmark was designed: it names the tested sectors, platforms, time window and three types of buyer question. But GrackerAI sells AI visibility and optimisation software, and the report says its visibility scores use a proprietary methodology and should be treated as directional benchmarks.
Publicly available material does not provide every submitted prompt, vendor-level answer, repeated run, country or account setting, or the exact classification rules needed to reproduce the 73% result independently. The researcher’s own recommendation to invest in particular content tactics is also not evidence that those tactics caused the reported omissions. A company announcement confirms what its research says; it is not independent validation of the findings.
The March 2026 core update is not an explanation
The earlier article connected the visibility gap with Google’s March 2026 core update. That chronology does not work. The GrackerAI research period ended in January 2026, while Google records its March core update as running from March 27 to April 8, 2026. The later update cannot have caused results collected before it began. Nor would a Google Search ranking announcement, by itself, establish why ChatGPT did or did not cite a cybersecurity vendor.
No diagnosed schema or authorship failure
The report does not isolate missing structured data, author biographies, off-site mentions or product-page wording as the cause of the 73% figure. Those elements can be checked when investigating a specific website, but they must not be presented as experimentally established causes of this benchmark result.
Google’s official guidance for its own generative AI search features says structured data is not required and no special schema.org type is necessary. Those rules concern Google Search, not ChatGPT. Accurate markup can still serve supported Google Search features, as MOCOBIN explains in its schema markup guide. It is not a guaranteed route into vendor recommendations.
How a security vendor should test its own position
A commercial benchmark can justify a closer look. It cannot tell an individual vendor whether a content change is needed. I would start with the questions a potential customer needs answered, then check whether the brand appears accurately under comparable conditions. The distinction between discovery and recommendation matters more than a single visibility score.
- Define the product category and buying task. Use the company’s actual capabilities and buyer requirements to choose a small set of category, comparison and constraint-based questions. Do not test the company against categories it does not serve.
- Record the environment. Keep each prompt’s wording, platform, date, language, market and available model or retrieval settings. Note whether the answer was generated with a fresh session and whether the interface shows clickable sources.
- Count separate outcomes. Record whether the brand is named, recommended, linked to directly, supported by a third-party citation, described accurately or absent. Do not combine those outcomes into one unexplained percentage.
- Repeat the sample. Compare multiple runs under similar conditions, and retain the responses where the service and its terms permit. Report the number of successful answers tested and failures; do not turn this sample into a claim about all users of the platform.
- Investigate only recurring gaps. Compare the missing categories with the information on the site, credible external sources and actual product fit before planning content changes.
For a simple internal metric, define the sampled vendor-citation rate as the number of successful prompt responses citing the chosen domain divided by the number of successfully captured responses in that same platform and test set. A separate brand-mention rate needs its own definition. Neither estimates how many real buyers saw the brand or whether an answer influenced a sale.
What would justify changing the website?
The response depends on what the check uncovers. If AI answers repeatedly put a vendor in the wrong security category, review the product pages and public descriptions first. Confirm that supported integrations, relevant environments, limitations and certifications are stated accurately. Where claims affect regulated or security-sensitive decisions, supporting evidence matters more than more promotional copy.
If the brand is named accurately but the assistant cites an independent review rather than the company’s website, that is not necessarily a visibility failure. Check whether the cited review is current and whether the product information it repeats is accurate. Do not manufacture endorsements or assume that more structured data will change citation selection.
When a genuinely relevant vendor is absent from repeated category questions, missing or unclear evidence is one hypothesis to investigate. Relevant independent coverage, genuine product documentation and verifiable customer evidence may help readers evaluate it. They are not a guaranteed recipe for placement in an AI-generated list. MOCOBIN’s broader AI visibility strategies examines possible editorial and off-site actions; this study alone does not validate any specific tactic’s effectiveness.
It would be premature to publish dozens of near-identical ‘best security tool’ pages, add unsupported schema types or buy a new monitoring platform solely because the report says 73%. The stronger reason to act is a repeatable and relevant omission, supported by a record of what was tested and by evidence of an actual information gap.
Report ChatGPT results separately from Google visibility
In June 2026, Google announced dedicated generative AI performance reporting in Search Console, with worldwide availability recorded by August 31. Those reports show impressions in supported Google AI features, including AI Overviews and AI Mode. They do not report how often a brand was recommended in ChatGPT. Google’s rollout announcement is the appropriate source for that distinction.
Keep Search Console AI impressions, sampled ChatGPT brand mentions, sampled linked citations, identifiable referral visits and qualified leads as different measures. A citation in a research prompt does not prove a visit, and a visit does not establish that an AI-generated shortlist caused it.
The best use of GrackerAI’s finding is as a reason to check a specific category-recommendation workflow, not as a universal forecast for cybersecurity vendors. If repeated tests reveal a real omission, identify the missing information and its owner. If the brand is accurately represented but rarely linked, report that narrower result rather than calling it invisible.
References
- GrackerAI: The State of AI Search Visibility in Cybersecurity, 2026 Benchmark Report (February 2026)
- GrackerAI: Full benchmark PDF, methodology and limitations (2026)
- Google Search Status Dashboard: March 2026 core update dates
- Google Search Central: Generative AI search optimisation guidance
- Google Search Central: Generative AI performance reports, June 2026 and August availability update









