How the SeoGauge Score Is Calculated

The exact checks, weights and pass thresholds behind SeoGauge's SEO, AEO and GEO scores — 35 scored checks, each category weighted to 100.

By the SeoGauge team · Published October 4, 2026 · Updated October 8, 2026

How is the overall score calculated?

SeoGauge runs 35 scored checks. Each check earns full, half or zero credit, each category score is the weighted share of credit earned (0–100), and the overall score is the plain average of the SEO, AEO and GEO scores.

Free reports show an SEO & AEO score: the average of those two categories. Pro unlocks the GEO category and the overall average of all three. Compare plans.

Scores of 75 or more are Strong, 50–74 Need work, and below 50 are Weak. The weights in every category add up to exactly 100, so a check's weight is also its maximum contribution to that category score.

These are readiness scores from website signals, not measured rankings, traffic, or AI citations. Search Console performance is shown separately in your private website project.

Where do the weights come from?

The weights follow published evidence rather than taste: Lighthouse's SEO audits for crawlability and on-page basics, Google's guidance on AI features, and the GEO study by Aggarwal et al. (KDD 2024), which found quotations, statistics and cited sources raised a page's share of AI answers most.

"There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."

"Through rigorous evaluation, we demonstrate that GEO can boost visibility by up to 40% in generative engine responses."

That is why SeoGauge weights being indexable, readable without JavaScript and easy to quote above special files and markup. The 40% figure is a benchmark result on test queries, not a traffic promise.

Two checks are preconditions: being indexable for SEO and letting AI crawlers in for GEO. As in Lighthouse, failing one costs about a third of that category instead of zeroing it, so the rest of the score still shows how ready the page is once it is fixed. Every weight change is tested against saved copies of pages with known outcomes: encyclopedia, reference and government health pages must score at least 60 on AEO, and placeholder, thin and JavaScript-only pages must stay at or below 35.

What changed in scoring version 5?

• AEO now rewards answer paragraphs under any heading, a lead answer, and lists and tables. FAQ schema dropped from 20 to 5 points and question headings from 20 to 10, because answer pages rarely phrase headings as questions.
• GEO adds quotations and cited sources. llms.txt dropped from 15 to 5 points; AI crawler access rose to 30.
• SEO adds crawlable links, descriptive link text and hreflang. Indexability rose to 30 points. Speed and security checks share 16 points, and render-blocking and request limits were relaxed to match real sites. Analytics is shown but not scored.

What changed in scoring version 6?

• Canonical tag now checks where it points, as Lighthouse does: full credit only for a valid absolute URL pointing at the page itself, half for a relative URL or one pointing elsewhere.
• Lead answer skips forms, buttons and short microcopy between the H1 and the first real paragraph, so tool homepages are judged on their actual text.
• Domain AI visibility: when live web search is measured it now carries 60% of the AI visibility score, up from 40%, because model memory can't know brands newer than its training data.

Reports made before this change keep their original scores and are marked with their scoring version; comparisons between versions show a warning.

How does the domain rating work?

A domain check discovers pages from sitemaps and a limited crawl, then audits up to 18 representative pages. Homepage, pricing, and product pages carry more weight. PageSpeed runs on up to 3 key pages and AI content review on up to 6 pages. The report shows the audited sample and how much check weight was measurable.

• SEO: 60% sampled page scores and 40% site-wide checks.
• AEO: 70% sampled page scores and 30% site-wide checks.
• GEO: 35% sampled pages, 25% site-wide checks, 30% API recommendation probes, and 10% authority proxies. If probes are unavailable, the remaining parts are normalized to 100%.

Site-wide checks cover duplicate titles/descriptions, canonicals, sitemap health, host redirects, links, language/security headers, answer-paragraph, schema and Q&A coverage, AI crawler policy, llms.txt quality, entity markup, and raw HTML content. They describe the measured sample, not every page on the website.

AI recommendation probes ask API models buyer questions using their own knowledge. On domain ratings started from a Pro account, the same questions are also asked to an API model with live web search; we record whether it recommends the site and whether one of the site's pages was among the search results it read. When live search is measured it makes up 60% of the AI visibility score (35% recommendations, 25% pages read), because model memory can't know brands newer than its training data; model memory makes up the remaining 40%. Neither probe captures answers from the consumer ChatGPT, Gemini or Perplexity apps. Authority proxies use registration age and Wikipedia/Wikidata presence; they are not backlink authority. Unavailable authority lookups may affect the heuristic score.

How do rescans and comparisons work?

Choose a fresh check after making a fix to bypass the 24-hour result cache. This uses the same daily allowance as a new scan. Saving a reused result retains its original measurement time. An in-progress domain rating is reused to avoid duplicate work.

Projects compare two distinct completed domain checks. Score deltas appear only when scoring versions, model configuration, audited pages, and measured checks match. Individual checks can still be compared when those totals differ. A previously failing check is verified only when the same page and check now pass; missing or unmeasured results remain unverified. AI probe questions may change between runs, so these comparisons do not establish an AI visibility trend.

To follow a trend, Pro projects can track up to 10 fixed buyer questions. Each check asks every question once from model memory and once with live web search and stores whether the site was recommended, its position, and the names recommended instead. Tracked questions are re-asked weekly. Pro projects can also turn on a weekly domain re-check, which builds the score history shown on the project page.

How are ready-made fixes written?

For some findings (llms.txt, FAQ schema, Organization schema, robots.txt rules for AI crawlers) SeoGauge can write the fix. It re-reads the live page, asks an AI model only for wording drawn from that page, and assembles the file itself: llms.txt links are limited to links found on the page, and structured data is built as valid JSON-LD. Generated text can still be wrong, so read it before publishing. The AI analysis of a page is reused when the page content has not changed.

Which data sources does SeoGauge use?

• Your page's raw HTML, fetched like a crawler (no JavaScript), plus robots.txt, sitemap.xml and llms.txt from your domain.
• Google PageSpeed Insights for Largest Contentful Paint and mobile performance, using real-user Chrome data when Google has it.
• One AI review of the page text for concise answers, entity clarity and citable facts. The AI also writes the page-specific advice in each fix.

The AI crawlers checked in robots.txt are: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot, Google-Extended, CCBot. Results are cached for 24 hours, so a repeat scan of the same page returns the same report.