DABYTE DATA DESK
How the DABYTE AI Visibility Index is measured
Answer. A fixed panel of 16 category buyer prompts is submitted to 3 AI engines (openai, perplexity, gemini). For each answer we record whether a tracked brand is named. Share of answer = the share of panel prompts naming the brand, averaged across engines with equal weights. Re-measured weekly. Paid placement never affects a score.
Current release
Rules
- The prompt panel is fixed between releases; any change is versioned in the changelog.
- A brand counts as named when the answer references the brand or its documented aliases.
- Scores are engine-weighted, then rounded to one decimal.
- Paid placement never affects a score. Commercial products of VECTORY (profiles, reports, measurement) live outside the scoring fields; no brand in the index has paid to appear in it.
- Engines that cannot be measured in a release are excluded and weights re-normalised — never filled with an assumed value.
- Resolution: one mention on one engine = 2.1 points. Each prompt is run once per engine per release, so a movement of a single step is within the noise of LLM non-determinism. Such movements stay visible in the table but are never reported as «gainers» or «losers» — only changes larger than one step are.
- Visibility is not market share, revenue or quality. It measures one thing: whether an engine names the brand when a buyer asks the category question.
- Cadence, disclosed: re-measurement runs weekly. The 9 releases already in the archive were taken 3–9 days apart while the panel and engine set were being settled, so intervals computed from the archive will not all be seven days. Every release carries its own date and panel version, and deltas are never computed across a panel change.
Prompt panel
Panel version 3. A share of answer is a share of this list, so the list is frozen between releases and every measurement records the version that produced it. We do not compute a change across two different panel versions — see the archive, where such rows read «new panel» instead of a movement that did not happen.
best AI writing assistant for teamstop project management software for startupsbest CRM for small businessbest AI coding assistanttop no-code app builderbest collaborative whiteboard toolbest customer support helpdesk softwaretop data analytics platform for product teamsbest workflow automation toolbest AI note-taking apptop design tool for product teamsbest knowledge base softwarebest team communication and chat app for workbest marketing automation platform for b2bbest applicant tracking system for hiringbest tool to build internal apps and admin panels
Engines we run, and the one we dropped
This release runs 3 engines: ChatGPT, Perplexity, Gemini. A third, Google Gemini, was measured and then excluded, and saying so here matters more than the tidiness of a two-engine table. Run through our relay it returned zero mentions for every brand across the whole panel — a result that is not a finding about the market but a failure of the measurement. Publishing it would have put a column of zeros next to real numbers; dropping it silently would have been worse, because the engine count is part of what the score means.
So the rule is written into the code rather than left to judgement: a build in which any engine returns zero across the entire panel refuses to publish at all. Weights are re-normalised across the engines that did answer, never filled with an assumed value. Gemini returns to the index when the relay is fixed and it produces a measurement we can defend — and its return will be a new panel version, not a quiet edit.
Ownership disclosure
DABYTE is published by VECTORY, an AI-visibility
company. This is disclosed here, in the footer of every page and in /humans.txt, because a
measurement is only useful if you know who ran it. VECTORY sells measurement and advisory services;
it does not sell positions in this index.
Reuse
Data is published under CC BY 4.0. Cite as:
DABYTE AI Visibility Index — SaaS & AI Tools, 2026-09-07. dabyte.ai
Machine copies: /api/aiv.json,
/aiv.csv, /index.md,
/llms.txt.