OLAB AssistHow OLAB Assist works
A plain account of what we read, how results are scored, what monitoring catches and how AI answers are measured. Every result in the product links back to the evidence behind it.
What your dashboard looks like
The OLAB Assist dashboard for olabassist.com, adapted from: the score, issues to fix, how often AI assistants name the brand, the intent and tone of each answer, and the alerts that go to your team.
See it run
A walkthrough on olabassist.com with its real audit; the AI answers and the drop in week 5 are example data. Use the steps or press play.
Enter a domain
Type any public domain. No sign-up and nothing to install on the site.
Video walkthrough: under 2 minutes
The whole product on olabassist.com: free audit, evidence and fixes, answers from six AI assistants, competitors, trends and exports, integrations, sharing and plans. Account data in the video is sample data.
What we read
For each audit we request public addresses on the domain: the homepage, /robots.txt, /llms.txt and /sitemap.xml. Requests identify themselves as OLABAssistBot, follow at most four redirects, stop after eight seconds and read at most 1.5 MB. We then request the homepage once each as OAI-SearchBot, Claude-SearchBot and PerplexityBot, to see whether a firewall treats them differently.
- No login and no access to your systems is needed. We only see what any crawler sees.
- JavaScript is not executed, on purpose: most AI crawlers do not run it either, so this shows what they can read.
- Private and local addresses are refused, and only ports 80 and 443 are used.
The checks
Grouped by the question they answer. Each finding in a report shows the request we made, the status code we got and the code we read.
| Group | Question it answers | Signals |
|---|---|---|
| SEO | Can search engines index and describe the page? | title, meta description, robots meta, H1, canonical, sitemap |
| GEO | Can AI systems reach the site and identify the business? | AI search crawlers in robots.txt, a live test as those crawlers, training crawlers, Organization schema |
| AEO | Is the content shaped for direct answers? | Product, service and article schema, question headings, Open Graph |
| AIO | Can a crawler that does not run JavaScript read it? | robots.txt validity, text in the HTML source, page language, valid JSON-LD |
How the score works
Every check has a weight from 1 to 5 based on its impact. A pass earns the full weight, a warning earns half and a failure earns nothing. Notes, such as a missing llms.txt or blocked training crawlers, carry no weight: they are choices or optional extras, not problems. A group score is the earned weight divided by the possible weight, times 100. The overall score is the average of the four groups.
Below 50 is poor, 50 to 79 needs work, 80 and above is good.
Example: a group with weights 3, 2 and 1 where the first check passes, the second warns and the third fails earns 3 + 1 + 0 = 4 out of 6, so it scores 67.
Evidence you can check
Results are only useful if you can verify them. Each finding shows the exact URL we requested, the status code, and the raw tag or file lines we based the result on. If something is missing, we show the part of the page we searched. Reports can be shared with a link or saved as PDF.
Weekly monitoring
Sites in your account are scanned again every week, and you can run scans on demand within your plan. Each scan is compared with the previous one:
- Worse: a check that passed now warns or fails, for example after a deploy changed robots.txt.
- Fixed: a problem from the last scan is resolved.
- New: a check that did not apply before now does.
The score history and every change stay in your account, so a drop can be traced to a date.
AI answer tracking
You add the questions your customers ask. Each one is sent to ChatGPT, Gemini, Perplexity and Claude through their official APIs with live web search turned on, and searched on Google in your language and country to capture AI Overviews and AI Mode. Then it is checked again every week.
- Mentioned: your brand or domain appears in the answer text. The rank is its position among the brands you track.
- Cited: your site is one of the sources the assistant links to.
- Share of voice: across the latest answers, how often each tracked brand is named.
- Every result keeps the full answer, the sources, the assistant, the model and the time it was asked.
- When Google shows no AI Overview for a question, it is marked as such and not counted as a miss.
What this does not claim
- A high score means the technical basics are in place. It does not guarantee rankings or AI citations.
- AI answers vary between runs, users and locations. We measure a repeatable sample, not every possible answer.
- API answers can differ slightly from the consumer apps, which add personal context.
Privacy
We only read public pages. Your sites, prompts and results are visible only to your account, and prompts are sent to AI providers without your name or email.