How to read your audit findings, and what to fix first
Short answer: The SearchScore AI Search (GEO) score is a 0 to 100 readiness benchmark built from a heuristic scan of your own website: it shows how ready the site is for AI to find, read and use. It is not a measurement of how often ChatGPT or any other assistant recommends your business. The score weights eight categories and places the result in one of five bands. The findings underneath tell you what the audit detected and which changes may be worth making. Use the score to understand your website’s condition, and the Tracker or your own checks to measure AI answers.
A single number is good for one thing: noticing that something has changed. It’s much less useful for deciding what to fix, because very different websites can arrive at the same total. This guide explains what goes into the score, what it leaves out, and how to turn a list of findings into a sensible order of work.
What does the free audit actually check?
One complete SearchScore audit checks 300+ signals across three disciplines: AI visibility (GEO), search (SEO) and conversion (CRO). They’re grouped into 23 scoring categories: 8 for GEO, 8 for SEO and 7 for CRO. Each discipline gets its own 0 to 100 score.
The GEO score on its own draws on 150+ of those signals across its 8 categories. The audit reads them from the live site: the pages themselves, the HTTP responses, robots.txt, the sitemap, llms.txt where one exists, and some public signals about the brand. In the words of our methodology page: “If a signal cannot be verified, we score it conservatively.”
How is the score built?
This section is the short version. For the definition, see what a GEO score is, and for the detail, how the readiness score is calculated.
The GEO score is a weighted sum of eight categories. Each category is scored 0 to 100, then multiplied by its weight. These are the weights the scorer applies (checked against the scorer on 12 November 2026):
| Category | Weight | Examples of what it checks |
|---|---|---|
| E-E-A-T Content | 24% | Author bylines and bios, publication and update dates, direct answers, data points, contact and about information |
| AI Citability | 18% | AI crawler access in robots.txt (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot and others), quotable statistics with sources, llms.txt, HTTPS |
| AI Platform Readiness | 12% | Bingbot and PerplexityBot access, IndexNow, Bing verification, answer-first structure, FAQ, HowTo, QA or Speakable markup |
| Structured Data | 12% | JSON-LD present, Organisation, Article, Person, FAQPage and Product schema, completeness of fields |
| Technical | 12% | robots.txt, sitemap, HTTPS, server-rendered content, canonical tags, meta descriptions |
| Brand Authority | 10% | Social proof, LinkedIn and YouTube presence, consistent name, address and phone, other public profiles |
| Topical Authority | 8% | Content hubs, deep content, headings, internal linking, breadcrumbs, llms.txt |
| Platform Optimisation | 4% | Open Graph and social card tags, RSS feed, video content |
The examples are a selection, not the full list of checks. The audit currently still gives some credit for FAQ and HowTo markup, which no longer earns a Google rich result (see schema markup in 2026), and for Speakable markup, a beta feature Google uses for news read aloud. Treat findings about those types as low priority.
E-E-A-T Content carries the most weight, not AI Citability, so thin, anonymous content costs more than awkward formatting, and a gain in a heavily weighted category moves the total more than the same gain in a lightly weighted one. The weights describe how we score readiness. We haven’t published evidence that they measure business impact or the effect on whether assistants mention you, so don’t read them that way.
All three scores (GEO, SEO and CRO) share the same five bands. Only the labels differ.
| Range | GEO label | SEO and CRO label |
|---|---|---|
| 80 to 100 | AI-Ready | Excellent |
| 70 to 79 | Strong | Strong |
| 40 to 69 | Emerging | Moderate |
| 20 to 39 | Low Visibility | Weak |
| 0 to 19 | Invisible | Critical |
Above AI-Ready sits a Verified tier at a GEO score of 90 or more.
How rare the top bands are (a dated snapshot, not a live count). In the SearchScore corpus snapshot of 27 July 2026, published with the Q3 2026 SAVI report, 216 of 1,000,000+ sites analysed scored 80 or more (about 1 in 4,600), 0 reached Verified, and the highest GEO score was 86.6. Those figures are frozen until the next SAVI edition; they describe that snapshot, not today’s corpus.
The band labels describe the website against the signals we check. “Emerging” means the site has some of the foundations and is missing others. It isn’t a statement about how often an assistant names the business.
What the score does not measure
It doesn’t ask any AI assistant anything. The free audit never sends a question to ChatGPT, Gemini, Claude, Perplexity, Grok or DeepSeek. It reads your site and scores what it finds. Asking the assistants is what the paid Tracker does every week, across those six engines. The free AI recommendation check does it once, for one question.
It isn’t a prediction of how often you’ll be cited. Our methodology page puts it this way: the GEO score “is a readiness benchmark to improve against, not a prediction of how often you will be cited.” Whether an assistant names you also depends on things outside your site: what other websites say about you, how many competitors answer the same question well, and how each assistant chooses its sources.
It isn’t your Google ranking or your traffic. The SEO score checks search-ranking signals on the site; it doesn’t report where you rank. For rankings, traffic and AI referrals, use Search Console and your analytics.
It isn’t a verdict on your business. A brilliant local firm with an old website can score low. The score describes the website, not the service behind it.
It’s a snapshot. It reflects the site at the moment of the audit.
Which number is which?
Several numbers get confused with each other. They measure different things:
| Number | What it measures | Where it comes from |
|---|---|---|
| AI Search (GEO) score | How ready your website is for AI to find, read and use, against the signals we check | The free audit, from your own site |
| Tracker AI Visibility Score | How you appear in the assistants’ answers to a fixed set of questions | The paid Tracker, from the assistants’ answers |
| Mention rate | How often an assistant names your business in its answers | The Tracker, or your own manual checks |
| Citation rate | How often an assistant links to your site or names you as its source | The Tracker |
| Google rank | Where a page appears in Google’s results for a query | Search Console or a rank tracker |
| Traffic | Visits to your site, including any referred by an assistant | Your analytics |
Don’t read the Tracker’s AI Visibility Score against the GEO bands: the band names belong to the GEO score only. And when a report gives you an AI number, check whether it counts mentions or citations before comparing it with anything. The Tracker’s headline figure is a mention rate, and it’s labelled that way.
A worked example: a fictional 62/100 audit
Fictional example. “Example Plumbing” and every score below are invented to show how to read an audit. They are not a real customer’s results, and the category scores are not drawn from our data.
Example Plumbing’s audit comes back at 62/100, Emerging. Here’s how the total is built:
| Category | Category score | Weight | Contribution to the total |
|---|---|---|---|
| E-E-A-T Content | 55 | 24% | 13.2 |
| AI Citability | 70 | 18% | 12.6 |
| AI Platform Readiness | 60 | 12% | 7.2 |
| Structured Data | 40 | 12% | 4.8 |
| Technical | 85 | 12% | 10.2 |
| Brand Authority | 66 | 10% | 6.6 |
| Topical Authority | 70 | 8% | 5.6 |
| Platform Optimisation | 45 | 4% | 1.8 |
| Total | 62.0 |
Among its findings are these five. The priority labels are our suggested reading, not fields the audit produces:
| Finding | Category | Our suggested priority | Effort | In your control? | What to recheck afterwards |
|---|---|---|---|---|---|
| PerplexityBot is disallowed in robots.txt | AI Citability and AI Platform Readiness | Blocker | Minutes | Yes, unless your CDN or host sets the rule | robots.txt, then re-run the audit |
| Service pages have no author or business owner named | E-E-A-T Content | High | An afternoon | Yes | The pages themselves, then the audit |
| No Organisation schema on the home page | Structured Data | High | An hour | Yes | View source for application/ld+json, then the audit |
| No publication or update dates on guides | E-E-A-T Content | Medium | An hour or two | Yes | The guide pages |
| No Open Graph image on key pages | Platform Optimisation | Low | Under an hour | Yes | A social preview tool |
How to read it.
- The blocker comes first, even though the score barely shows it. The GEO score has no special rule for blockers: a crawler blocked in robots.txt costs only the points of the checks it fails, here in two categories, and doesn’t cap the total. So a site can sit at 62 with a crawler locked out. Fix that before anything else. If the rule is there on purpose, that’s your call, but make it knowingly.
- Then work by weight. The two E-E-A-T Content findings sit in the 24% category. The Open Graph finding sits in the 4% category. Both are worth doing; one is worth doing first.
- Check what you control. Some rules are set by a host, CDN or plugin rather than by you. Find out who owns the setting before you plan the work.
- Separate fixes from tips. Some checks can’t be seen from outside the site, such as Bing verification done by DNS record. The Fix-List shows those as tips rather than counting them as fixes, so check them yourself before spending time on them.
- Re-audit, then measure separately. A re-run shows whether the audit now detects the change. It doesn’t show whether any assistant’s answers changed. For that, ask the assistants the questions your customers ask before choosing, or use the Tracker.
Does the audit check llms.txt?
Yes, as a small part of two categories: AI Citability and Topical Authority each include a check for a valid llms.txt file. We don’t treat it as a requirement of any AI engine, because it isn’t one. Google’s guide to optimising for generative AI features says Google Search itself doesn’t use llms.txt files and that maintaining one “will neither harm nor help your site’s visibility or rankings in Google Search”, while noting it’s fine to keep one for other services that use it. No major AI search engine has confirmed reading one. Treat it as housekeeping, and put it behind every blocker and every higher-weight finding.
Why we built it this way
A score that claimed to predict citations would need evidence that it does, and we’d rather tell you exactly what it measures than overstate it. What a site-side audit can do well is check, consistently and at scale, the things you control: whether assistants can get in, whether they can understand what you do, and whether your pages give them something worth repeating. That’s what the score summarises, and it’s what the findings tell you how to fix.
Run a free SearchScore audit, skim the total, then spend your time on the findings.
Sources
SearchScore code and data checked 12 November 2026.
- SearchScore, Methodology
- SearchScore, What is a GEO score?
- SearchScore, AI Search Tracker
- SearchScore, Q3 2026 SAVI report (source of the dated AI-Ready snapshot)
- Google Search Central, Optimizing your website for generative AI features on Google Search (llms.txt)
Frequently asked questions
Does the free SearchScore audit ask ChatGPT about my business?
No. The free audit is a heuristic scan of your own website: crawler access, content structure, structured data, entity and authority signals and more. It doesn't send any question to ChatGPT or any other assistant. Asking the assistants is what the paid Tracker does every week, across ChatGPT, Gemini, Claude, Perplexity, Grok and DeepSeek. The free AI recommendation check does it once, for one question.
Does a higher GEO score mean I will be mentioned more often by AI?
Not necessarily. The GEO score is a readiness benchmark to improve against, not a prediction of how often you'll be cited. It tells you how well your site is set up to be found, understood and quoted. Whether assistants actually name you also depends on things outside your site, and that has to be measured by asking them.
Why did two sites with the same score get different advice?
Because the score is a weighted total of eight categories. Two sites can reach the same number with very different strengths and gaps. The findings list what each site is actually missing, which is why they, not the total, should decide what you fix first.
Does the audit check llms.txt?
Yes. A valid llms.txt file is one check inside AI Citability and one inside Topical Authority, a small part of each. Google says its Search features don't use llms.txt, and no major AI search engine has confirmed reading one, so treat it as low-priority housekeeping rather than a requirement.
My score went up after a fix. Does that mean AI will recommend me more?
It means the audit now detects the signal you fixed. It doesn't tell you whether any assistant's answers changed. To find that out, ask the assistants the questions your customers ask before choosing, or track it over time with the Tracker.