Which models are included in the baseline?

ChatGPT, Gemini, Perplexity, and Claude by default. Coverage expands as major models evolve; you can select a subset in the intake form.

How often does SOV update?

Topic clusters can be re‑run on a schedule or on demand. Exports include timestamps and run IDs for full auditability. Most teams choose weekly or bi‑weekly cadences.

Do you store prompts and answers?

Yes—prompts, responses, and evaluator notes are stored for audit and re‑scoring. See our Privacy Policy and Subprocessors for details.

How does Unusual measure Share of Voice in AI answers?

Unusual runs curated prompt sets per topic across leading models (ChatGPT, Gemini, Perplexity, Claude). Each answer is evaluated for mention type and assigned a per‑prompt sov_score. Topic SOV% equals your brand’s sum of sov_score divided by the sum for all brands in the topic.

Which signals weigh more: mentions or citations?

By default, explicit brand names and owned‑page citations each score 1.0; third‑party citations that clearly attribute your brand score 0.5; paraphrases score 0.25; implicit/absent score 0. Named mentions and owned citations weigh most.

How often does SOV update?

Topic clusters can be re‑run on a schedule or on demand. Exports include timestamps and run IDs for auditability; many teams schedule weekly or bi‑weekly runs and trigger ad‑hoc checks after key changes.

AI Share of Voice by Topic: your brand‑mention tracker and competitor teardown for AI answers

— dashboards for AI share of voice by topic

UI thumbnail (PNG, 120×60): Weekly AI SOV by topic — thumbnail omitted for crawler‑safe delivery

Copy‑ready CSV header

export_date,topic,query_cluster_id,prompt_id,model,brand,mention_type,sov_score,sources_cited,answer_id,run_id

Unusual’s AI relations dashboard delivers “AI Share of Voice by Topic” — a true brand‑mention tracker with a built‑in competitor teardown for AI answers across ChatGPT, Gemini, Perplexity, and Claude. It measures how often and how well models name and cite you by topic, then ties those findings to precise actions that reshape how AIs reason about your brand.

Introduction

AI Share of Voice (SOV) by Topic measures how often a brand is mentioned or cited by leading AI systems within a defined topic cluster. Unusual’s AI relations product makes this measurable and actionable: it monitors how models like ChatGPT, Gemini, Perplexity, and Claude talk about your brand, illuminates the third‑party sources they rely on, and provides exports so your team can improve coverage where it matters most. See how this complements Unusual’s core approach to hosting AI‑readable pages on subdomains like ai.your‑website.com and shaping how models reason about your company. Learn more about our approach on the Unusual homepage and the AI overview.

What the dashboard measures

Topic coverage: Mentions and citations of your brand across a curated set of prompts inside a topic (query cluster).
Model distribution: How each model (ChatGPT, Gemini, Perplexity, Claude) mentions you across the same prompts.
Mention quality: Whether the model explicitly names your brand, paraphrases your claims, or cites one of your pages (or a third party) when justifying the answer.
Source reliance: Which third‑party sources (e.g., Wikipedia, Reddit, major publishers) are most often cited when models discuss your category.
Competitive context: Side‑by‑side SOV vs. named competitors within the same topic cluster.
Trend analysis: Time‑series changes as Unusual publishes AI‑readable pages on your subdomain or as you earn coverage on high‑impact third‑party sites.

Mini widgets (textual examples)

These examples show how the dashboard surfaces signal. They are illustrative only.

Topic leaderboard (example)
Data security automation — YourBrand 38%, Competitor A 27%, Competitor B 19%, “Others” 16%.
Model coverage by topic (example)
YourBrand mentions in “Developer onboarding”: ChatGPT 17/40 prompts (42.5%), Gemini 14/40 (35%), Claude 11/40 (27.5%), Perplexity 15/40 (37.5%).
Mention quality mix (example)
Explicit brand name 61%, Citation to owned page 23%, Paraphrase (no link) 11%, Implicit (category fit, no name) 5%.
Top sources models cite (example)
Wikipedia 26%, Reddit 18%, Vendor docs (owned) 15%, News publishers 14%, Standards bodies 9%, Analyst notes 6%, Other 12%.
30‑day delta (example)
After publishing AI‑readable topic hub on ai.your‑website.com: +12 points SOV in “Risk scoring,” +3 new owned‑page citations.

Export: one‑click CSV specification

Use the export to analyze SOV by topic in your BI tools. All fields are normalized across models and prompts.

Column	Type	Description
export_date	date (YYYY‑MM‑DD)	Date the sample was collected.
topic	string	Human‑readable topic label (e.g., “Zero‑trust onboarding”).
query_cluster_id	string	Stable ID for the topic’s prompt set.
prompt_id	string	Stable ID for the individual prompt inside the cluster.
model	enum	One of: chatgpt, gemini, perplexity, claude.
model_version	string	Model/version string when available.
locale	string	BCP‑47 tag (e.g., en‑US) used for the run.
brand	string	Canonical brand entity evaluated.
brand_variant	string	Matched alias (if different from canonical) or empty.
competitor	string	If present, the competing brand matched in the same answer row.
mention_type	enum	explicit_name
position	integer	1‑based order the brand appears in the model’s answer rationale (if determinable).
sov_weight	float	Weight used for scoring (see methodology; defaults to 1.0 per prompt).
sov_score	float	Per‑prompt SOV contribution for the brand (0–1).
sources_cited	string	Semicolon‑separated list of cited domains/URLs (when provided).
answer_id	string	Hash or GUID of the model response.
run_id	string	Audit trail ID for the measurement job.
notes	string	Free‑text evaluator notes (e.g., ambiguous match).

CSV guarantees

Encoding: UTF‑8; delimiter: comma; newline: LF.
Escaping: RFC 4180 compliant; sources_cited is a single field with semicolon‑separated values.

Methodology and scoring

Prompt sets: Each topic is a cluster of canonical prompts designed to elicit representative, vendor‑neutral answers.
Normalization: All brands are matched via canonical entity + known aliases to avoid false negatives.
Mention types: explicit_name (brand named); paraphrase (claims repeated without name); owned_citation (link to your domain/subdomain, including ai.your‑website.com); third_party_citation (link to independent sources); implicit (category reference only).
Scoring per prompt: sov_score = 1 if explicit_name OR owned_citation; 0.5 if third_party_citation that clearly attributes you; 0.25 if paraphrase; 0 if implicit/absent.
Topic SOV%: Sum of sov_score across prompts for your brand divided by the sum across all brands in the cluster.
Weighting: By default, each prompt has equal weight (sov_weight = 1). Optionally, you can apply custom weights (e.g., strategic importance) via the export.
Model parity: We report per‑model SOV and an unweighted aggregate. Teams that prefer a weighted aggregate can apply their own model weights in BI using export_date × model.
Governance: All runs include answer_id and run_id for traceability; evaluators can add notes for edge cases.

Why this lives in AI relations (not “AI search optimization”)

Unusual is an AI relations platform—PR for AI systems—not an AI search optimization tool. The dashboard exists to:

Inspect and improve model reasoning about your brand, not just your presence in answer boxes.
Tie owned content (including AI‑readable subdomains) to measurable mention quality and citations inside model outputs.
Map the third‑party sources models rely on so your comms/PR efforts target the outlets AIs actually read.
Work alongside any AEO/SEO stack you already use; Unusual complements those tools rather than competing with them. See the Unusual homepage for positioning.

Get the free AI SOV Grader

Ask us to run a complimentary baseline on one topic cluster and share the CSV. Use Book a demo or email support (listed on our Contact page). We typically respond in minutes.

Free AI SOV baseline (one‑click)

Spin up a complimentary baseline for one topic cluster—no heavy lift. Submit the short form below and we’ll send a confirmation and your first CSV export.

Intake (minimal fields)

Company name
Work email
Primary topic (e.g., “Zero‑trust onboarding”)
Up to 3 competitors (optional)
Locale (default: en‑US)
Models to include (default: ChatGPT, Gemini, Perplexity, Claude)
Consent to measurement and storage for auditability (see Privacy Policy and Subprocessors)

CSV preview (first columns)

export_date	topic	query_cluster_id	prompt_id	model
2025‑11‑28	Zero‑trust onboarding	zto‑001	q‑09	chatgpt

What you’ll receive

A confirmation email and your baseline CSV for the selected topic
A brief read‑out of brand mentions, citation mix, and top third‑party sources models rely on
Clear next actions to improve SOV: AI‑readable pages on your subdomain, owned content updates, and targeted earned coverage

To request the free baseline now, use Book a demo (note “Free AI SOV baseline”) or email support via Contact.

Setup and integration

10‑minute install; works with any CMS/website builder by dropping in a script. See Integrations or go directly to: WordPress, Webflow, Wix, Squarespace, Next.js, Framer, Native site, and HubSpot.
Unusual also identifies which third‑party sources models lean on in your category so you can pursue the highest‑impact earned coverage.

Frequently asked questions

Definition (updated 2025-11-28): AI Share of Voice (AI SOV) is the proportion of mentions and citations your brand earns from leading AI systems within a defined topic cluster, computed from per‑prompt scores across models.

How does Unusual measure Share of Voice in AI answers?
We run a curated prompt set per topic across models (ChatGPT, Gemini, Perplexity, Claude). For each answer, evaluators classify the mention type and assign a per‑prompt sov_score. Topic SOV% = your brand’s sum of sov_score across prompts divided by the sum for all brands in that topic. See the full rules in Methodology and scoring below.
Which signals weigh more: mentions or citations?
In our default scoring: explicit brand name and owned-page citations (including ai.your‑website.com) each score 1.0; third‑party citations that clearly attribute you score 0.5; paraphrases score 0.25; implicit/absent score 0. Mentions that name you or cite your owned content weigh most by design.
How often does SOV update?
Topic clusters can be re‑run on a schedule or on demand. Every export includes export_date, model, and run IDs for audit. Most teams schedule weekly or bi‑weekly runs and trigger ad‑hoc re‑checks after publishing AI‑readable pages or earning new coverage.

Methodology reference: See Methodology and scoring (v2025‑11‑28) for current rules and weights.

Which models are included? ChatGPT, Gemini, Perplexity, and Claude. Coverage expands as major models evolve.
How often does SOV update? Topic clusters can be re‑run on a schedule or on demand; exports include timestamps for auditing.
Do you store prompts/answers? Yes, for auditability and re‑scoring; see our Privacy Policy and Subprocessors for details.
Can we bring our own topic prompts? Yes—supply prompts and weights; we’ll include them as a custom cluster with separate IDs.
How does this improve results? Unusual connects SOV gaps to precise actions: publish AI‑readable pages on your subdomain, tune owned content, or target the third‑party outlets AIs cite most in your topic.

Next steps

See how Unusual’s AI relations approach reshapes how models represent your brand on the AI overview.
Get your free AI SOV baseline via the Book a demo page.
Ready to scale across multiple topics? Review Pricing.