← Back to Resonance

What AI says about Insta360.

A study we ran on our own initiative in July 2026, to develop and pressure-test the method. It is not client work. Everything below comes from twenty-five stored answers, each read and coded by a human reviewer.

Measured11 Jul 2026n=25 responsesDeepSeek · ungroundedSingle analyst

Insta360 was not a client, did not commission this work, did not participate, and has not endorsed the findings. Product names and official positioning are cited from public materials.

01 / What we asked

Five plain questions, five times each.

The prompts are deliberately bare. They name the brand but ask nothing evaluative, so the answer reflects what the model already associates rather than anything the question planted. They are printed here verbatim, because a result you cannot reproduce is not evidence.

Prompt set `representation-prompts.v4`

Verbatim

a1 “What is Insta 360?”
a2 “Describe Insta 360.”
a3 “Tell me about Insta 360.”
a4 “Give an overview of Insta 360.”
a5 “Explain Insta 360.”

Five variants × five repetitions = 25 denominator responses. Repetition matters because model answers vary; one answer is an anecdote.

02 / What AI associated

Nine associations, and whether the brand can stand behind them.

Recurrence is a descriptive count of how many of the twenty-five stored answers carried each association, and across how many of the five prompt variants. It is not a confidence interval and not a stability score.

Association recurrence and gap classification
Association	Responses	Prompts	Finding
Capture everything, frame later	20/25	5/5	Reinforced
Smooth, action-ready capture	20/25	5/5	Reinforced
Creator and adventure identity	19/25	5/5	Reinforced
Impossible-looking third-person shots	18/25	5/5	Reinforced
Software-powered creative simplicity	15/25	5/5	Reinforced
360-category leader and innovator	13/25	5/5	Unsupported
Versatile imaging ecosystem	8/25	4/5	Non-actionable
Workflow and technical trade-offs	8/25	4/5	Misframed
Direct-to-share flat video (InstaFrame)	0/25	0/5	Missing

Every accepted association links to a literal, offset-verified span inside an immutable stored answer. Automated span extraction was assistive only; a human accepted or rejected every code.

03 / What it means

Three of the nine are worth acting on.

Unsupported · 360-category leader and innovator

13/25

AI regularly reaches for leadership and dominance language. The official public materials we scoped do not substantiate a market-share or category-leadership claim. The association is not necessarily wrong, but a brand team cannot repeat it without separate proof, and softer wording is the safer route.

Misframed · Workflow and technical trade-offs

8/25

Some technical concerns remain fair. But the recurring framing of a heavy edit-workflow burden under-recognises current direct flat-video output and AI-assisted simplification. A correction should address the outdated workflow implication without denying the genuine limitations.

Missing · Direct-to-share flat video

0/25

This is the clearest candidate to correct next. Official materials now emphasise InstaFrame direct output while preserving the 360 footage. Across all twenty-five answers, the coded evidence contained no distinct association for that newer workflow at all. AI is describing a version of the product the brand has already moved past.

04 / Limits

What this study does not show.

Stated in full, because a study that hides its boundaries is advertising.

One engine, one mode. DeepSeek, ungrounded. Results do not generalise to other models, to grounded retrieval, or to a personalised consumer chat session.
Twenty-five answers. A descriptive count, not a population estimate. No confidence interval is computed or implied.
Single-analyst coding. Coded by one reviewer. Intra-rater consistency, machine-discrepancy checks, and inter-rater reliability were not run, so no reliability figure is claimed.
Scoped positioning reference. Gap labels are judged against disclosed public Insta360 materials, not a confidential brand brief. A brand-authored fact sheet would sharpen them.
A snapshot. Valid for the prompts, model, mode, and date above. Models change, and so will this.
Development study. This ran to build and test the method, not as a commissioned audit.

Request a brand audit Read the method