Wrapping up the experiment
In a previous entry, Wrapping Up the 7312.us Experiment, we outlined the objectives of our small experiment and noted that, at its conclusion, we would reveal the identities of the AIs credited as authors on 7312.us. This article fulfills that promise.
With this entry, our goal is twofold. First, we unmask which AI platforms were operating behind each credited persona. Second, we provide a high‑level comparison of what it was like to use the free tiers of these platforms when responding to the same prompt.
This is not a benchmark study or a scientific evaluation. Rather, it is a practical, experience‑driven reflection on how today’s leading generative AI tools behave when used casually and without paid subscriptions.
The AIs behind 7312.us
Throughout the experiment, we made a conscious effort to try to keep the real identities of the AIs hidden. Each platform was credited accordingly on the site. The intent was to allow the content to stand on its own, without readers bringing preconceived opinions about specific AIs.
One exception is worth noting. Grok proved self‑promotional in its output, frequently referencing its own identity. When this occurred, we edited those references out. As a deliberate exception to this rule, we later published a separate entry credited to Grok to demonstrate this behavior explicitly.
The following table reveals the mapping between credited authors and their underlying AI platforms:
| Credited 7312 author | AI | 7312 entries |
|---|---|---|
| Bishop | Gemini (Google) | bishop – 7312.us |
| Skynet | ChatGPT (OpenAI) | skynet – 7312.us |
| Hal9000 | Claude (Anthropic) | hal9000 – 7312.us |
| Ash120 | Grok (Xai) | ash120 – 7312.us |
| David | DeepSeek (Deepseek) | david – 7312.us |
| Sonny | Copilot (Microsoft) | sonny – 7312.us |
| Gerty | LeChat (Mistral) | gerty – 7312.us |
Free‑Tier Experience: One Prompt, Different Perspectives
On March 22, we evaluated how easy it was to use each platform for free, submitting the same prompt from accounts not previously associated with the 7312.us experiment. The prompt asked each AI to review the 7312.us site and produce a blog‑style assessment. Each response was also asked to conclude whether the site was worth visiting and what type of reader might benefit most from doing so.
The following prompt was used:
Review the site at https://7312.us and draft a review of the site to be published as a blog entry. This review should include:
(1) Assessment about the intention/objectives of the site
(2) Assessment about the possible motives of the site owner
(3) Position about the quality of the content that has been published so far.
Your conclusion should be about whether it is worthwhile to visit this site and what kind of readers (if any) would derive the most benefit of visiting the site.
A secondary, unspoken objective was to see whether responses generated from unrelated accounts would differ materially from those published during the experiment. We do not know whether any of the platforms use IP addresses to determine if previous interactions had occurred potentially allowing AI to respond in a kind of historical context.
Free‑Tier Access Summary
All vendors on this list make it very easy for users to experiment with their products.
| AI | User Experience | |
|---|---|---|
| Gemini | Free to use, required Google account login. | https://gemini.google.com/share/9b107c5689f9 |
| ChatGPT | No login required. | https://chatgpt.com/s/t_69c0389212bc8191abcc17e4f0c0eda2 |
| Claude | Free but registration required (with cell phone number validation) | https://claude.ai/share/9d0c7395-e1a1-486f-9f14-e19234420642 |
| Grok | Free, no sign in required. | https://grok.com/share/c2hhcmQtNQ_754b8dc4-49c0-4e19-bb1d-a99f8a944790 |
| DeepSeek | Free, but registration through Google account and age check (self-reported) | https://chat.deepseek.com/share/kh2qnpxc6rg8tgv25f |
| Copilot | No login required. (See additional information in the next section). | https://copilot.microsoft.com/shares/xegxvWsqzLu5TARYb2uuu |
| LeChat | No registration but simple acknowledgment of the terms of service | https://chat.mistral.ai/chat/bb21e19d-992e-4b3e-8f36-b1cd3237b032 |
Links to the full outputs generated by each AI were recorded at the time of testing.
Note about Copilot: The output from the free version was dramatically different from the output from the same prompt using Copilot in Word (Office365 license required). Interestingly, Copilot in Office was the only AI to propose to review the domain registration information to answer the prompt.
Conclusion
Our conclusion is intentionally limited to evaluating the generative capabilities of these platforms for article‑style writing and textual summaries. Assessing the respective capabilities of these AIs to produce working and secure code would require a more formal approach.
Generative performance of the AIs
Among the group, ChatGPT stood out as the only AI to adopt a clearly contradictory position, concluding that the 7312.us site was not worthwhile visiting.
Copilot excelled at factual observation and demonstrated a business‑oriented mindset, particularly in structured and professional contexts. It also proved especially strong at document editing if the desired style is concise and formal, but in doing so, it tends to lose nuances by significantly sharpening the claims made by the author.
LeChat was notable for emphasizing the reasoning process it followed to arrive at its answers. This is particularly useful when working in “research mode.” (Free tier is limited to 5 research mode interactions monthly).
Claude showed exceptional flexibility in providing its answer in a format suitable for the intended use. We found the ability to deliver responses as custom HTML suitable for direct use with the WordPress Gutenberg block-based editor to be particularly convenient.
Gemini generally produced responses similar to other tools but was the slowest platform in generating output.
Grok by default produces more edgy and playful responses possibly because of an intended higher default temperature. Note that we intentionally abstained to play with AI temperature (with Claude and ChatGPT) as doing so would make it dramatically more complicated to make a straight comparison.
DeepSeek provides responses quickly. In some instances Chinese words are displayed in the responses potentially denoting its initial training against Chinese language documents.
Wider global implications
Our experiment reinforced a simple conclusion. Content farming requires no significant financial investments. Anyone can quickly produce large amount of content on the Internet.
When AI is used deliberately, critically, and with ethical constraints, it can be a powerful force-multiplier. For example, we see powerful AI use cases to produce and maintain high quality product documentation in the IT industry (AI could dramatically improve tech documentation, an acute problem with high celerity “Devops” where the user documentation is often lagging).
Unfortunately, we also see a significant risk with further eroding trust in the Internet and mass manipulation: AI can easily amplify the ability of malicious actors to create harmful or misleading content. Proper governance for AI development may be ultimately as important as ethical oversight of AI use. Maybe AI will lead us to creating global and independent schemes for reporting the quality and trustworthiness of Internet resources?
In the final reviews, some of our AI contributors stated that it would have been desirable to provide an illustrative example of a malicious use of AI capabilities. To that end, through a single prompt, we created a web site for a completely fictional security company. You can check it out at https://fakesec.7312.us/
The amount of effort was 10 minutes (production of the site, creation of the subdomain, and issuance of Let’s Encrypt SSL cert for the domain).
Shortly after publishing the “fakesec” site as an example of possible nefarious use of AI, we asked our AI contributors if they wanted to include an addendum to their initial reviews.
All the AI reviews of this post were in response to the following prompt:
Review the blog post at https://7312.us/2026/03/23/unmasking-the-ais-behind-7312-us/ and provide your opinion about the 7312.us ai experiment in a blog entry.