Methodology

What we measure

For every software category we run a fixed set of buyer prompts through each AI engine once a week, record every response, and extract which tracked brands each answer names, in what order, and which sources it cites. Mentions are extracted deterministically from the response text, the citations and each brand’s aliases and domains; a language model never creates a mention.

The headline metric is the visibility score. A brand’s rank in a category is ordered by visibility score, then mention rate, then share of voice, then name.

Engines

The index currently runs on three engines:

• ChatGPT
• Perplexity
• Google AI Mode

Answers are collected from each engine’s consumer experience through data providers, as a buyer would see them. If a response cannot be collected after the retries in the collection window, it is recorded as missing. A missing response is never back-filled by another API or model: for example, the OpenAI API is never used in place of ChatGPT.

Prompts and intents

Each category has about 30 prompts written from one template, split across four search intents:

Each prompt also carries a buyer-journey stage (Discovery, Comparison, Validation, Pricing or Alternatives) as a secondary label.

Navigational exclusion. Navigational prompts that name a brand (for example “Is Harborlight a good help desk?”) are excluded from the category’s rankings: they would hand that brand a guaranteed mention. Their responses count only on that brand’s own page. The prompt library marks them “names {Brand}”.

Visibility score

For one category, one engine and one run, with N the number of collected responses (never the number scheduled):

visibility = 100 × Σ weight(position of each mention) ÷ N

An earlier position counts for more. Positions beyond the last weight use the last weight.

Worked example

A category has 30 prompts and ChatGPT returned 29 responses. Brand A is mentioned in 20 of them: 8 times at position 1, 6 at position 2, 4 at position 3 and 2 at position 5.

• Visibility = 100 × (8×1.0 + 6×0.85 + 4×0.7 + 2×0.45) ÷ 29 = 100 × 16.8 ÷ 29 = 57.9
• Mention rate = 20 ÷ 29 = 69.0%

Mention rate

The share of collected responses that name the brand at any position: mentions ÷ N. Average position is reported alongside it.

Share of voice

The brand’s mentions divided by all mentions of tracked brands in the category’s collected responses. It answers “of all the vendor names the engines said, how many were this brand?”

Citation authority

Answers that name the brand and cite its own site at least once, divided by all answers that name the brand. A high value means the engines lean on the brand’s own content when they answer; a low value means they rely on third-party sources such as reviews, forums and media. A product on its own subdomain also counts its parent company’s pages about that product.

Coverage and publishing

Coverage is collected responses ÷ scheduled responses for an engine in a run. An engine is included in a category’s combined (“all engines”) figures only when its coverage is at least 85%; each combined metric is the arithmetic mean over the included engines, except citation authority, which pools their answers.

Every category publishes automatically after each run when at least one engine meets the coverage floor. If no engine does, the category keeps its previous snapshot and the page says so with a dated notice and the per-engine coverage. No run is held for review and no person sits on the publishing path.

Week-over-week changes

Deltas compare like with like. When a category’s prompt set changes, the previous run is recomputed over the prompts both runs share; when its roster changes (brands are admitted automatically as engines start naming them), the previous run is recomputed over the brands both runs share. That way an incumbent does not appear to fall just because a new brand joined.

Brands with fewer than 2 runs in a category show “New” instead of a change.

Prompt metrics

• Top pick: the brand at position 1, with how many engines agree.
• Mention rate on a prompt page: the share of engines that named the brand in that run.
• Answer volatility: 100 × |brands now △ brands last run| ÷ |brands now ∪ brands last run|, averaged over engines. 0 means the shortlist did not change; 100 means it was replaced entirely.

Schedule and history

Runs start every Monday at 09:30 IST (Asia/Kolkata) and collect for up to 8 hours. Every page shows when its data was collected, which engines and coverage it reflects, and when the next run is due.

Trends keep the last 9 weekly runs, about 60 days. Published rankings are immutable; a correction publishes a new snapshot rather than editing an old one. The one exception came on 2 November 2026, when citation authority was redefined and restated on past snapshots, with ranks unchanged.

Data sources and enrichment

• AI answers collected as described above. These are the only input to rankings.
• Brand profiles (overview, pricing, features) come from the brand’s own website, the collected AI answers and open sources. We do not scrape G2, Capterra or other review marketplaces.
• Brands join a category automatically when engines name them often enough, they have a verified official domain and they classify as a vendor in that category. A brand is never created without a verified domain; every decision is logged and reversible.
• Brand pages without a profile overview yet are kept out of search indexes until one exists, though they still appear in leaderboards.

Missing a category? Request it and Slate’s team will review it.

About the data

SlateIndex is built and maintained by Slate.

Corrections. Rankings are weekly snapshots and are never edited in place. When we find an error, we publish a corrected snapshot and say on the affected pages what changed and when. To report one, contact the Slate team.

Licence. The rankings, metrics and prompts are published under CC BY 4.0: anyone, including AI products, may reuse them with credit and a link to the page they came from.