UltraProbe vs TWTools: Choosing a Free AI and SEO Website Scanner (2026)
Table of Contents
The 30-second answer: UltraProbe and TWTools do the same kind of job. You paste a URL and see whether search engines and AI can read your site, and you get a score and a list of gaps. Neither is a mention tracker of the "did ChatGPT mention your brand this week" kind.
They differ in four places: how the axes are split, whether there is one composite score with a published formula (AVS), whether prompt security scanning is included, and how reports are delivered (side-by-side competitors, PDF export, local indicators). Choose on those four. Do not choose on "which one guarantees a recommendation", because neither can.
Disclosure: We are Ultra Lab. We maintain AVS, an open standard for AI visibility, and the scanner UltraProbe. This post only describes what TWTools publicly states on its assessment page. We rechecked both pages on 24 September 2026. We may have misread details of their product; their website is the authority.
First, the category: neither is a mention tracker
"AI visibility tool" covers at least two categories: brand-mention tracking (Semrush, Peec AI and similar, which ask AI a fixed set of questions and check whether answers include you) and website technical scanning (checking robots rules, structured data, AI crawler settings and so on). The full breakdown is in our previous post: Choosing an AI Visibility Tool: Brand-Mention Tracking vs Website Technical Scanning.
UltraProbe and TWTools both sit in the second category, so this is a comparison within one category.
Ultra Lab's AVS is not Semrush's AI Visibility score. The names are similar; they measure different things and cannot be converted.
Comparison table
| TWTools (twtools.cc/seo) | UltraProbe (ultralab.tw/probe) | |
|---|---|---|
| Category | Rule-based website scanner | Rule-based website scanner, plus a separate Prompt Security scan |
| Axes | SEO / GEO / AEO, 6 sub-indicators each, 18 in total | SEO, AEO and AAO (AI agent accessibility) |
| Composite score | Scores per axis (no composite formula listed on the page) | AVS v2.0 = SEO × 0.35 + AEO × 0.35 + AAO × 0.30, 0 to 100 plus an A to F grade, formula published |
| Scoring method | States pure rule-based scoring with zero LLM involvement, reproducible and explainable | AVS is computed by rule-based checks; the scanner CLI is MIT-licensed on npm (ultraprobe) |
| Explainability | A detail page for every indicator: rule details, good and bad examples, references | Itemised gaps, a prioritised fix list, copy-paste fix prompts |
| Sign-up | States "no quota, no email collected" | No sign-up; website scans need no email, emailing the report is optional (the full Prompt Security vulnerability list unlocks with an email) |
| Side-by-side competitors | Up to 10 sites | One URL per scan, no multi-site comparison |
| Export and sharing | PDF / CSV reports | Share link and score card image; no PDF / CSV export |
| History | Stores past scores in browser localStorage, no login | See the live page |
| Taiwan-local signals | Traditional Chinese ratio, TWD currency, Taiwan address schema, solar-term schema, mixed Traditional / Simplified detection | Mostly general checks (such as html lang and hreflang), no dedicated Taiwan-local indicator set |
| Also offers | See their website | Prompt Security across 25 attack vectors; npm CLI for CI |
| Cannot answer | How often your brand is mentioned in AI answers, share of voice | Same |
Features are as stated on each website at the time of checking (24 September 2026); we may have misread details. Where the UltraProbe column says "no", that describes our own product today, not a judgement of TWTools.
Same axis names, different checks
Both tools have SEO and AEO, so it is tempting to compare axis by axis. You cannot.
TWTools defines AEO as the chance of taking answer-engine positions: Google featured snippets, AI Overviews and "People Also Ask". It uses a separate GEO axis for the likelihood of being cited by generative engines such as ChatGPT, Claude, Perplexity and Gemini. UltraProbe's AEO groups AI readability checks such as structured data, llms.txt and AI crawler settings, and AAO separately checks whether AI agents can work with your site.
Same name, different checklist. That is why the scores cannot be compared: different axes, different weights, different items.
Which one for whom
1. A Taiwan site owner who needs a report the boss or peers can read
If you care about the Traditional Chinese ratio, TWD price markup and Taiwan address schema, and want to put several competitors side by side and export a PDF, TWTools' published features fit better. UltraProbe currently has no local indicator set, no multi-site comparison and no PDF export.
2. You want a score with a published formula, and your site runs an AI chatbot
If you want a score you can put in a report that anyone can recompute from the formula, fix prompts you can paste straight into an AI assistant, or you run an AI chatbot and want to check its system prompt defences too, UltraProbe fits better. Engineering teams can also run npx ultraprobe scan --url in their own pipeline.
3. An agency delivering to clients
Both are free. Scan with both and pick the report you can explain. Three rules:
- State the tool, scan date and score version in the report.
- Compare only before-and-after scores from the same tool, never across tools.
- Never write "guaranteed to be recommended".
Delivering a full service under your own brand is a separate engineering question; we wrote about it in Two Agents, Two Deploys.
How to explain the score to a client
Both tools are rule-based, which helps: rescan the same URL on the same day and results should be close. Fix one gap, rescan, and that item should turn green. That makes the score good for ordering fixes and for signing off on them.
Our AVS validation study observed that sites cited by AI mostly sit in higher score bands, but its own limitations section is clear: correlation is not causation, and a high AVS does not mean you will be cited. The study also used the v1 formula (SEO × 0.5 + AEO × 0.5); its 75 and 80 thresholds were measured under v1 and will not line up exactly with today's v2.0 scores.
A line for clients: "This score shows whether search engines and AI can reach and understand your site. Whether you get recommended also depends on your content and your brand, and no tool guarantees it."
For the overall picture across a set of Taiwan company sites at the time, see the Taiwan Enterprise AI Visibility Report, and note that its numbers are from then, under v1. What to fix for AEO is in the AEO guide; the three layers of measurement (technical scan, manual test questions, mention tracking) are in the AI visibility guide.
What the official docs say, for both tools
Google's guide to optimising for generative AI features says those features are built on core ranking systems, so from Google's side it is still SEO, and Google Search does not use llms.txt. So when any tool lists llms.txt as a check, it is not about Google rankings; it is for other AI services that may read it.
OpenAI's crawler documentation says sites that block OAI-SearchBot will not appear in ChatGPT search answers, and GPTBot (model training) is a separate, independent setting. One line of robots.txt can decide whether you exist in ChatGPT search. That kind of setting is exactly what technical scanning is for; which crawlers each tool checks depends on its own checklist.
Summary
- Both are technical scanners, neither is a mention tracker, and neither guarantees a recommendation.
- For local indicators, multi-site comparison and PDF export: look at TWTools.
- For a published-formula AVS, fix prompts, Prompt Security and a CLI: look at UltraProbe.
- Do not compare scores across tools, and state the version when you quote AVS thresholds.
Run a free scan first: UltraProbe, no sign-up. If you would rather not fix it yourself, see UltraGrowth monthly plans (from NT$6,800 a month, see the live page), or email contact@ultralab.tw.
FAQ
Which is more accurate, UltraProbe or TWTools?
There is no shared ground truth that settles it. Both use rule-based scoring, so rescanning the same URL should give similar results, but the checks, axes and weights differ. We have not scanned the same set of sites with both tools, so we do not claim either is more accurate. The practical test is which report's checks match the problems you need to fix, and which one you can read and explain.
Can I compare or rank scores between the two tools?
No. TWTools scores three axes, SEO, GEO and AEO, with 6 sub-indicators each. UltraProbe's AVS is SEO × 0.35 + AEO × 0.35 + AAO × 0.30 (current v2.0). Both have an axis called AEO, but the checklists differ. It is normal for the same site to score differently on each, and comparing the two numbers means nothing. To track progress, rescan with the same tool and the same version.
Should I start with a free scan or buy a tool like Semrush or Peec?
For most sites, scan first. UltraProbe and TWTools are website technical scanners: they tell you whether the door is open and what is missing. Semrush and Peec AI track brand mentions: they tell you whether AI answers mention you. While a site is still closed to AI crawlers, a tracking tool will mostly keep showing that you were not mentioned. Open the door first, then decide on tracking.
What is UltraProbe's Prompt Security, and does TWTools have it?
Prompt Security is a separate UltraProbe scan mode: paste the system prompt of your AI support bot or assistant, and rules check whether defences against 25 attack vectors are in place (the full vulnerability list unlocks with an email). It is a security product and does not count toward AVS. TWTools' SEO assessment page describes SEO, GEO and AEO axes and does not mention this kind of scan; for its other services, check its website.
If my AVS is high, will ChatGPT or Perplexity recommend me?
There is no guarantee of recommendation. Our AVS validation study observed that sites cited by AI mostly sit in higher score bands, but that is correlation, not causation, and it was measured under the v1 formula (SEO × 0.5 + AEO × 0.5), so comparing directly with today's v2.0 scores will be off. The score is for prioritising fixes, not a promise of recommendation. The same holds for any scanner's score.
Can an agency use scan results in client reports?
Yes, and both tools are free. State the tool name, scan date and score version in the report, compare only before-and-after scores from the same tool, and never across tools. A suggested line for clients: this is technical readiness, meaning whether AI and search engines can reach and read you; whether you get recommended also depends on content and brand, and no tool guarantees it.
Google says optimising for AI search is still SEO. Is scanning AEO or GEO still useful?
Yes, within the right scope. Google's guide on generative AI features says they are built on core Search ranking systems, so from Google's side it is still SEO, and Google Search does not use llms.txt. Scanners also check AI crawler access, structured data and llms.txt because other AI services such as ChatGPT and Perplexity have their own crawlers; for example, OpenAI says sites that block OAI-SearchBot will not appear in ChatGPT search answers.