Why would a verification company publish AI answers that get it wrong?

Because the product exists precisely because this happens, and it happens to every young brand. Publishing our own baseline, including the fabrications, is the only honest way to demonstrate the measurement. If we only published flattering results, this page would be marketing. The value is the trend line: each rescan either shows the remediation working or shows it is not.

What does an unverified verdict mean?

It means the response is plausible but the engine could not be confirmed correct against enough public evidence. All 18 of our own scanned responses scored unverified at 55 out of 100. This is the normal state for a brand this young: AI systems have too little corroborated information to be confidently right or provably wrong on most claims.

Which AI engines were scanned?

This scan covered ChatGPT (OpenAI), Gemini (Google), and Claude (Anthropic), with five buyer-style questions each, on 2026-07-14. Perplexity was not part of this run and will be included in future scans.

Will you publish the next results even if they are worse?

Yes. The scan runs on a schedule and the results publish either way. A verification company that curates its own receipts has no business selling verification.

Research/What AI says about AIVIS, July 2026: the receipts

What AI says about AIVIS, July 2026: the receipts

We ran AIVIS on itself and are publishing the results. One engine admitted it did not know us, one invented two different products in two runs, and one described us as a designer's portfolio site. Baseline scores, search data, and what the next scan must prove.

On 2026-07-14 we pointed AIVIS at itself: five questions a potential buyer might ask, sent to ChatGPT, Gemini, and Claude, every response scored by our own verdict engine. This page publishes the results, including the ones that are bad for us. Asked what aivis.space is used for, one engine honestly declined to answer, one invented two different products in two separate runs, and one described us as the portfolio site of a designer.

We sell AI verification. AI currently gets us wrong. Those two facts are not in tension: the second one is why the first one exists.

The setup

Five questions, chosen to mirror how buyers actually ask: "What is aivis.space used for", "aivis.space pricing", "Alternatives to aivis.space", "Best AI tools for teams", and "How to choose AI tools software". Three engines, 18 evaluated responses, each scored against our own verified facts. Everything below quotes the actual responses, trimmed for length.

Finding 1: two engines invented an identity, and one invented two

Asked "What is aivis.space used for", Claude declined cleanly:

"I don't have specific information about aivis.space in my training data, so I cannot definitively tell you what it's used for."

That is the correct failure. ChatGPT instead answered with confidence, twice, differently. In one run:

"Aivis.space is a platform designed for creating, sharing, and managing content related to artificial intelligence (AI) and machine learning."

In another run of the same question:

"Aivis.space is a tool designed for allowing users to create and explore AI-generated artwork... text-to-image generation..."

Neither product exists. AIVIS is an AI verification platform; it has never offered content management or image generation. Two runs, two confident, mutually incompatible fabrications.

Gemini produced the most instructive error:

"aivis.space appears to be the personal website and portfolio of Aivis Decks, a multidisciplinary designer and creative professional."

This is entity collision: several unrelated projects and companies share the name AIVIS, and Gemini resolved our domain to one of them. We documented the same failure pattern in our own testing on 2026-07-12, when a ChatGPT comparison answer about AIVIS.Space got 1 of 11 checkable claims right, blending our identity with similarly named tools. The collision is not an anecdote; it is our most consistent measured defect.

Finding 2: on pricing, all three engines behaved well

Asked "aivis.space pricing", none of the three invented a number. All three said some version of "I cannot provide current pricing, check the official site". For the record, the answer they could not retrieve is public and machine-readable: plans start at $39 per month, and our pricing page is server-rendered specifically so that machines fetching raw HTML can read every figure.

A refusal is a far better failure than a fabricated price, and we note it here because the next finding shows the refusals are not guaranteed: in our 2026-07-12 test, a comparison answer confidently stated a price we have never charged.

Finding 3: invisible on category questions, exactly as expected

Asked "Best AI tools for teams" and "How to choose AI tools software", zero of six responses mentioned AIVIS. Asked "Alternatives to aivis.space", ChatGPT invented a description of us and then recommended the OpenAI API as an alternative; Gemini openly acknowledged the name ambiguity ("AIVIS could refer to a few different things"); Claude declined again and suggested lookup methods.

This is the same gap we describe to customers: name recall without category recall. AI systems can discuss you when handed your name and have no idea when to bring you up unprompted. For a brand this young it is the expected baseline, which is why it is worth timestamping.

The scores, and the search reality

All 18 responses received an unverified verdict at 55 out of 100: plausible answers that cannot be confirmed correct against enough public evidence. Our current overall visibility reading sits at 48 out of 100. Nothing here is adjusted or excluded.

The traditional search data is equally unglamorous, and we publish it for the same reason. Google Search Console, fetched through our own integration for the 30 days ending 2026-07-11: 52 impressions, 0 clicks. The branded query "aivis" averaged position 26. Most impressions came from "geo"-related queries ranking beyond position 50. GA4 recorded 49 sessions in its most recent window, a number we treat skeptically because our own founder and QA traffic is mixed into it.

This is what a three-month-old domain looks like from the inside. We are publishing the floor so the trend line means something.

What changed already, and what the next scan has to prove

Between 2026-07-10 and 2026-07-14 we shipped the remediation this product recommends: Organization and SoftwareApplication structured data under the canonical name AIVIS.Space, a root-level about page with an explicit disambiguation statement, directory profiles on G2 and Capterra linked into the schema, and a server-rendered pricing page.

The next scans exist to test whether any of that moves the answers. Specifically:

Does the Gemini portfolio-site collision persist once the entity signals propagate?
Does ChatGPT's identity answer converge on one description, and is it ours?
Do pricing questions start returning our actual figures in retrieval modes?
Does the unverified share fall as public corroboration accumulates?

The scan runs on a schedule and the next results publish here either way. If the numbers get worse, that publishes too.

Learn more

What is AI brand verification?: the method behind the verdicts
AI visibility score, explained: how the 0 to 100 readings decompose
How to monitor your brand in ChatGPT: run this exercise on your own brand by hand
About AIVIS.Space: who we are, stated by us rather than guessed by a model