About llmeknow
llmeknow is the platform for understanding AI influence on your brand: it measures what ChatGPT, Claude, Gemini, Grok, Kimi, and other AI models say about a brand or subject, and how those answers shift by audience segment.
For brand teams, communications teams and researchers studying AI recommendations, misinformation and model safety.
Know what AI says about you
Which brands does AI recommend? Does it repeat a false claim about your organisation or give an outdated account of a public issue?
llmeknow helps you answer those questions across AI models and audience segments. Compare what the models say, read their cited sources and track whether their answers change over time.
The founders
llmeknow was founded by Aldu Cornelissen and Kyle Findlay. They also co-founded Murmur Intelligence, a Cape Town consultancy that studies online conversations and influence.
Dr. Aldu Cornelissen
Aldu is a data scientist specialising in social network analysis, the study of how people and information connect. He previously worked at Kantar and co-founded the Computational Social Science Group at Stellenbosch University.
Kyle Findlay
Kyle previously led a data science team in Kantar’s Global Innovations group. His work applied AI to consumer research, analysing social media and customer feedback to understand how people describe brands.
What llmeknow does
Compare AI models
Ask the same questions across models to see which brands they recommend and how they describe them. Compare mentions, sentiment and the attributes associated with each brand.
Compare audience segments
Test how answers change when you give AI different audience profiles. Compare each segment with a baseline that has no audience profile attached.
Track changes over time
Repeat a study in research waves. See whether your brand appears more often, whether its description changes and which sources the models cite.
The llmeknow research engine
Define what you want to extract from AI answers. Your entity classes can cover brands, people and places, or abstract concepts such as pain points, needs and values. You set the categories; there is no fixed list.
Watch an answer become data
Illustrative answer using fictional companies.
One answer, three entity classes extracted together. These are example classes; you define your own.
- Companies
- Attributes
- Pain points
Example answer
Cedar Bank is easy to use, and customers describe Cedar’s account setup as simple to use. Northstar Bank offers more branches, but reviews mention long waits and slow replies from support, as well as unexpected charges.
What happens next
The same answer contains company names, product qualities and customer frustrations. Watch them become structured results together.
Custom entity classes: define what to extract
Define your own entity classes to match your research question. Our extraction pipeline records their mentions, order and sentiment, turning free-form responses into data you can compare across models and audiences.
Normalisation: keep the same entity together
A company’s full name, abbreviation and spelling variants can otherwise split its results. llmeknow combines semantic clustering with AI-assisted review to group name variants under a consistent name.
Narrative taxonomies: measure the claims behind the words
Organise the claims you want to track into themes, topics and statements. A separate evaluation step scores how strongly each response supports or challenges each statement, distinguishing a balanced answer from one that never discusses the topic.
Trace the findings back to the answers
Compare entity mentions and narrative stance across models, audience segments and research waves. The original responses remain available so you can check the wording behind a finding.
Read our methodology and comparison with other tools for more detail.
What you can investigate
Brand visibility and reputation
Brand teams and agencies can compare which brands AI recommends and how it describes them. Communications teams can examine answers about a company or executive.
AI safety research
Researchers can review responses for stereotypes, agreement with false premises, or differences in when a model refuses to answer. Ask the same questions across models and simulated audience segments, then repeat the study to track changes in those behaviours.
Misinformation in AI answers
Check whether AI repeats a false claim about an organisation, person or public issue. Compare responses across models and audience segments, then inspect the cited sources to investigate where the claim appears.
Suspected AI poisoning
Look for repeated changes in claims, recommendations or cited sources when you suspect deliberate manipulation. Compare models and research waves to document what changed and identify answers that need further investigation.
Tracking corrections
After publishing a correction or updating your information, repeat the same questions to see whether AI answers change. Check whether an outdated claim persists and which sources are still cited.
Explore our public research case studies. The organisations and subjects studied are not necessarily customers.
How to get started
Set up your study
Choose the questions you want to ask, the AI models to test and the audience segments to compare.
Review the results
A research wave collects the answers for your study. Compare the results, read individual responses and share your findings in a campaign report.
Get help from the team
Book a demo to discuss what you want to measure. You can also send product or research questions through our contact page; we reply within one business day.
Key facts
- Product
- llmeknow
- Operator
- Murmur Intelligence (Pty) Ltd
- Type
- AI perception research platform
- Co-founders
- Dr. Aldu Cornelissen and Kyle Findlay
- Operator location
- Cape Town, South Africa
- Core offering
- Research into what AI models say about a brand or subject, and how answers shift by audience segment.
- Website
- llmeknow.com
- Communication
- Contact the team
Frequently asked questions
Who operates llmeknow?
llmeknow is operated by Murmur Intelligence (Pty) Ltd, based in Cape Town, South Africa.
Does this replace customer research?
No. llmeknow measures what AI says, not what your customers think. Audience segments are simulated profiles supplied to the models, not people taking part in a survey.
Can llmeknow change what AI says about us?
llmeknow helps you find missing or inaccurate information in AI answers and examine the sources behind it. It cannot control those answers or guarantee a recommendation.
How does llmeknow support AI safety studies?
Researchers define the questions and assess the responses against their own criteria. Results describe behaviour under the tested prompts and settings; llmeknow does not provide a validated safety score or certify that an AI system is safe.
Can llmeknow prove that AI has been poisoned?
No. AI poisoning involves deliberately manipulating information an AI system learns from or uses. llmeknow can help you examine suspicious outputs, but it cannot inspect training data or private assistant memory, establish intent, or prove that a change was caused by an attack.
How is the research carried out?
Each study asks a set of questions across your chosen models and audience segments. Repeated responses let you compare how often brands or claims appear; another research wave lets you track changes over time.