Measurement method
How GetSounding measures AI answer visibility
GetSounding runs one controlled set of unbranded buyer questions across the AI services included in a plan. For each usable answer, it separates three facts: whether the service cited the audited site, whether the answer named the brand, and how a confirmed brand mention was framed. Paid plans also check Google AI Overviews when available. Last reviewed 3 August 2026.
What each report headline means
- Search foundations
- A GetSounding score based on verifiable search foundations such as metadata, page structure, content, mobile setup and technical checks.
- Domain citation visibility
- The share of completed service-question checks that cited the audited hostname or an accepted subdomain. Failed checks are excluded.
- Brand mention rate
- The share of completed answers with available answer text that name the brand. A source link by itself is not a mention.
- Net sentiment
- The average framing of confirmed mentions from -100 to +100. A sentiment score appears once at least three confirmed mentions can be assessed.
- Answer structure
- A GetSounding checklist for relevant-question coverage, direct answers and extractable semantic structure. It does not predict whether a service will select the page.
- Across-service consistency
- The share of comparable questions where at least two services cited the audited hostname or an accepted subdomain. At least two services must return usable answers for a question to be comparable.
- Coverage
- The number of included AI services that returned at least one usable answer.
How SoundingScore works
A page audit measures one submitted page with fifteen buyer questions. SoundingScore measures brand visibility within a published fixed audit cohort. The ranking includes each audit target, plus brands found in another audit's answers or cited sources.
SoundingScore formula: 60% brand mention rate + 25% cited source reference rate + 15% official brand source citation rate.
Each component rate is rounded to four decimal places before weighting,
and the final score is rounded to two. The result is shown on a scale
from 0 to 100. Search and filters change which brands you see, not their
scores.
- Brand mention
- How often AI answers name the brand.
- Cited source reference
- How often returned sources clearly point to the brand.
- Official source citation
- How often an AI answer cites a known official brand site.
- Appearances
- The number of distinct AI answers that mention the brand or return a source that clearly points to it. A mention and source in the same answer count once.
The matrix plots distinct appearances from left to right and SoundingScore from bottom to top. Its dashed lines show the fixed full-corpus medians, so filters never change a brand's position.
Public data and privacy
Public SoundingScore data contains aggregate brand measurements. It does not include customer accounts, saved customer context, raw questions or raw answers.
What does an AI answer-visibility audit measure?
GetSounding counts a citation only when the audited hostname appears in a returned source URL. It counts a brand mention only when the recorded answer text refers to the audited brand or domain. Sentiment is assessed only around a confirmed mention; missing or uncertain evidence stays unavailable rather than being treated as neutral. The result is a controlled sample, not a census, search-volume estimate or promise about the next answer.
Models used for controlled checks
| Service | Guest and Free | Paid plans |
|---|---|---|
| Perplexity | Perplexity Sonar | Perplexity Sonar Pro |
| Gemini | Not included | Gemini 3.1 Flash Lite · web |
| OpenAI | Not included | GPT-5.1 · web |
| DeepSeek | Not included | DeepSeek V3.2 · web |
| Claude | Not included | Claude Sonnet 4.5 · web |
These are controlled, web-enabled model checks, not recordings of a visitor using each consumer app. Google AI Overview checks appear as their own report row because they observe a Google Search result.
How Google AI Overview checks work
Paid audits look for an AI Overview in a non-personalized Google.com search result in US English on desktop through an external search-data provider. The report shows whether an overview appeared, whether it linked to the audited site, and the source evidence returned.
If Google returns no AI Overview for a tested question, the report records no appearance. If the check fails or returns no usable evidence, it is marked not measured rather than negative.
Counting rules
- Citation
- A returned source URL uses the audited hostname or one of its real subdomains. Source order determines position. A model interpretation cannot create or rank a citation.
- Mention
- The recorded answer contains a confirmed reference to the audited brand or domain. A mention can occur with or without a citation.
- Sentiment
- Positive, neutral, negative or mixed framing around a confirmed mention. Low-confidence classifications remain unavailable.
- Not measured
- The service did not return a usable result. This is not evidence that the audited site was absent.
- Standard and custom questions
- Standard brand metrics use the unbranded question set. Custom questions are reported separately.
Repeat audits compare like with like
A trend includes only audits run with the same measurement method. Free uses one service while paid plans use five. The first paid audit therefore becomes a new baseline rather than being compared with a Free score.
The protocol
- Retrieve the submitted public page, metadata, headings, links, structured data and crawler rules.
- Confirm a bounded brand identity from the page and audited hostname.
- Generate exactly fifteen plausible, unbranded category and buyer-intent questions from that page.
- Send the same controlled set to the services included in the plan and retain the question, completion status and returned sources.
- Measure citation evidence and answer-text brand presence independently, then assess only confirmed mention contexts.
- Build scores and recommendations from the saved evidence, with unavailable checks excluded from the relevant denominator.
Guest and Free audits use Perplexity Sonar. Paid audits use Perplexity Sonar Pro, repeat the same questions across five AI services and include a Google AI Overview check when one is available. If no overview appears, GetSounding records no appearance. If a service fails or returns an unusable response, GetSounding marks the affected measurement unavailable without changing any independent result that was successfully observed.
Crawler access and visibility
Crawler rules can affect whether a service can retrieve or index a page. Source selection still depends on the provider. Search access, user-requested retrieval and model training can use different controls.
Source documentation
Limitations
- Answer engines are probabilistic and can change sources between identical runs.
- The generated questions are a controlled sample, not observed demand data.
- Service outages and missing source data reduce the usable denominator.
- GetSounding scores are not endorsed by any search or AI service.
- No schema, crawler rule, llms.txt file or page edit can guarantee ranking, citation or inclusion in Google AI Overviews.