Methodology
How the audit works, and what the sources are.
We separate deterministic checks from AI-assisted interpretation, and we name the source behind each finding. Data we cannot verify is shown as unavailable rather than guessed.
Deterministic technical checks
HTTP status codes, robots.txt directives, sitemap presence and validity, canonical tags, JSON-LD / structured data, meta robots and X-Robots-Tag headers. These are repeatable checks with no model in the loop; the same input produces the same finding.
Business identity and register checks
Legal name, contact details and applicable registers. Where relevant we cross-reference CQC (UK healthcare) and SRA (UK solicitors). If a register does not apply to your business, it is marked unavailable, not guessed.
Google Profile alignment
Primary category, service area, hours and website link retrieved via Google Places, cross-checked against the website. Findings describe observable mismatches, not assumed intent.
Directory presence
Searched via a directory lookup for the industry and locations that apply to your business. Directories that are not relevant to your sector are excluded rather than penalised.
AI-assisted interpretation
Perplexity is used for cited web answers. A Gemini-assisted pass is used to summarise the collected evidence into a report. Prompts and transcripts are stored with the report so you can see exactly what was asked and what was returned.
What we will not do
We do not predict Google rankings. We do not promise placement in any AI answer. We do not present fabricated statistics, testimonial carousels or “AI-powered” badges.
Handling missing data
If a source is unavailable, or a check cannot complete, the report shows it as unavailable and the score is marked provisional. We do not substitute unknown values with defaults that would flatter the score.
Variability of AI answers
AI assistants change their answers over time and between providers. Recommendation observations in the report are a snapshot at the time of the audit; a re-audit re-runs the same prompts against the current model output.