Varela InsightsSEO, AEO and GEO Audit: How We Measure Your Visibility in the AI Engines That Actually Cite
The Varela Insights audit measures the three layers of your company's digital visibility (technical SEO, AEO and GEO) with real citation measurement across the 5 families of AI engines that dominate the public Search Arena (GPT, Claude, Gemini, Grok and Perplexity, run in 7 configurations), a double-run rule at temperature 0, manual anti-homonym re-classification, a deterministic 9-criteria rubric, a census of local listings (Google, Bing, Apple) and raw evidence with a SHA-256 hash. It comes with an executive report, an actionable plan and the full JSONL of responses. The detail page is at varelainsights.com/auditoria and the complimentary audit runs free at varelainsights.com/geo.
Why do most "SEO for AI" audits fall short?
Because they measure too little and in ways you can't verify: a single run (which can be luck), no rule against homonyms (which confuses your brand with a similar company), no raw evidence (a summary you can't audit) and no look beyond your site (where the listings that map engines use live). This audit was born the other way around: it's the tool we use to measure our own brand and our clients' brands every two weeks, published as a methodology.
Which AI engines does it measure, exactly?
According to the public Search Arena (over 1.1 million votes), the top spots in the ranking are held by models from 5 families: GPT (OpenAI), Claude (Anthropic), Gemini (Google), Grok (xAI) and Perplexity (Sonar). Many "top 20 AI search engines" lists in circulation are duplicated models from those same families. Our battery runs 7 configurations of those 5 families, including Pro variants and search-enabled ones, with real buying questions from your industry, phrased the way your customer would ask them. Covering the families with real questions is worth more than listing twenty brands that share an index.
The 3 anti-false-positive rules
- Double run at temperature 0. Each question is run twice, identically; your brand only scores if it appears in both. A mention in a single run is noise, not citation.
- Manual anti-homonym re-classification. A real case we resolved: "Aceros y Metales Metro" versus "Aceros y Metales de Monterrey" are different companies; an automatic meter would have counted someone else's citation as your own. Every mention is classified by a human with the entity alongside.
- Evidence with a hash. Each run is saved as a raw response in JSONL with a SHA-256 hash. Any figure in the report traces back to its evidence; nothing is "trust us."
The 9-criteria rubric that scores every page on your site
| # | Criterion | Weight | Passes if… |
|---|---|---|---|
| 1 | Content without JavaScript (SSR/crawlable) | 15 | The full text is in the raw HTML the crawler sees. |
| 2 | Answer capsule after the H1 | 15 | The first 1-2 sentences answer the page's question, extractable in 40-50 words. |
| 3 | Question-style headings | 10 | ≥70% of the H2/H3s are questions or facets, and their first paragraph answers them. |
| 4 | Self-contained sections | 10 | Each section makes sense without reading the neighboring ones (a complete citable fragment). |
| 5 | Citable format | 10 | There is at least one list or comparison table in the body. |
| 6 | Density of cited sources | 15 | Figures or claims include attribution ("according to [primary source]"). According to Princeton's GEO study (Aggarwal et al., KDD 2024, arXiv:2311.09735), adding citations and statistics raised content visibility by up to 40%. |
| 7 | Visible Q&A in HTML | 10 | FAQs appear in the page body (not just in JSON-LD), and the first sentence answers the question. |
| 8 | Facet coverage | 10 | The 5-8 sub-questions behind your main query are answered on your site. |
| 9 | Freshness | 5 | dateModified is present and updated whenever the content is reviewed. |
What does the off-site census look at (and why is the audit incomplete without it)?
In map-type answers, the pin, phone number, hours, and stars come from local listings, not from your website. The census verifies your presence and NAP consistency (name, address, phone) across Google Business Profile, Bing Places, and Apple Business Connect, plus open signals like OpenStreetMap. Two technical reasons: Applebot is one of the most active crawlers on consumer sites on iPhone, and engines without their own business registration (DeepSeek, Qwen, Z.ai's GLM) have no sign-up form for businesses: they inherit what already exists in indexes like Bing's. A well-built Bing listing works for them too.
A typical audit vs. this audit
| Aspect | Typical audit | Varela Insights audit |
|---|---|---|
| AI engines measured | 1 (sometimes ChatGPT) | 5 families / 7 Search Arena configurations |
| Runs per question | 1 | 2 at temperature 0; counts only if it appears in both |
| Homonyms | Not checked | Manual re-classification per entity |
| Evidence | PDF summary | Raw JSONL with a SHA-256 hash per run |
| Local listings (GBP/Bing/Apple) | Not included | Full NAP census with status per channel |
| Site score | Generic checklist | Deterministic 9-criteria rubric (100 points) |
| Deliverable | Diagnostic | Report + prioritized action plan + evidence |
What exactly do I get?
- Executive report: how many questions AI names you in, who it recommends instead of you, rubric score per page, and the 5 GEO metrics (Citation Rate, Schema Coverage, Entity Anchoring, NER Accuracy, Drift).
- Prioritized action plan: what to fix first, with the rubric criterion or local listing that moves it.
- Complete evidence: a JSONL file with the verbatim answers from each engine, question by question, with an integrity hash.
- Live walkthrough: a 30-minute video call to walk through the findings with you.
How much does it cost and how does it start?
The complimentary audit (one question, 3 engines) is free and runs in ~25 seconds at varelainsights.com/geo, no sign-up. The full audit (a battery of questions for your industry, 7 configurations, rubric, off-site census, plan and evidence) is quoted after a no-cost 15-minute video call; typical delivery is 5 business days, CFDI 4.0. Demo cases of the method with anonymized data are shared on that call.
Who is behind the methodology?
Irving Varela: Doctorate in Administration (UANL), PMP, PMI-CPMAI #4391271 (certified 2026-05-26), founder of Varela Insights in Monterrey, Nuevo León. We apply the same measurement to our own brand and publish the result: in the September 2026 measurement, Varela Insights appeared in 8 of 9 runs for the question “who does GEO in Monterrey?” The test we ask of any auditor: you should be able to verify it yourself with a single question to AI.