Obsidian is a label for any instance where an AI chat system directly quotes or paraphrases a brand's owned content in its response
01What it is and how it works
When a large language model (LLM) receives a query, it searches its internal knowledge base and any indexed external sources. If the model finds a piece of content that matches a brand’s official page, press release, or product description, it may insert that snippet verbatim or in a close paraphrase. The platform tags that insertion as Obsidian so marketers can see when the model is using brand‑authored material instead of third‑party commentary. The tag is added at the generation layer, not in the source code of the model.
Obsidian means an AI answer that shows a brand's own words or data.
02What to do about it
1. Audit your brand’s top‑ranking pages for completeness and brand voice consistency. 2. Add structured data (e.g., Article or FAQPage schema) to help the model identify the right content. 3. Create a short “AI‑ready” summary for each key page and place it near the top of the HTML. 4. Set up a weekly alert in the monitoring dashboard to flag any new Obsidian instances. 5. If an instance is inaccurate, submit a correction request through the model provider’s feedback channel.
- Run a crawl of your site with Google Search Console and note pages that appear in the top 5 results.
- Add
meta name="robots" content="noai"only on pages you never want to be quoted.
03How it is measured or noticed
Our platform captures the raw LLM output, then runs a fuzzy‑match algorithm against your indexed brand assets. When the similarity score exceeds a preset threshold (default 85 %), the snippet is labeled Obsidian. The dashboard shows the query, the exact quoted text, the source URL, and a confidence score. You can also export the data as CSV for deeper analysis.
04Common mistakes
- Relying on generic meta descriptions alone; the model prefers richer content like full articles.
- Leaving duplicate content across subdomains; the model may pick the less‑optimal version.
- Assuming every Obsidian tag is a win; some quotes may be outdated or contain errors.
05Limits
Obsidian only appears when the model decides to surface brand‑authored text. If a brand’s content is behind a login wall, in a PDF, or blocked by robots.txt, the model cannot see it and will not generate an Obsidian tag. The term is also sometimes confused with “citation” – a citation may reference a source without quoting it, while Obsidian always includes a direct excerpt.
06Worked example
"When asked ‘What are the key benefits of Product X?’, the AI replied: ‘According to the official Product X page, the three main benefits are speed, reliability, and low cost.’ This excerpt was flagged as Obsidian because it matches the exact wording on the brand’s product page."
Frequently asked questions
How is Obsidian different from standard brand mentions in AI responses?
It is different because Obsidian refers specifically to instances where the AI directly quotes or paraphrases brand‑owned content, not just any reference. Standard mentions might be generic or derived from public knowledge, while Obsidian pulls from your indexed assets.
Should we try to prevent Obsidian occurrences, and what factors influence that decision?
It depends on your brand policy and legal requirements. If you need strict control over how proprietary wording is displayed, you may implement safeguards; otherwise, occasional Obsidian may be acceptable as long as it aligns with your usage guidelines.
How does the platform detect Obsidian instances in LLM output?
It works by capturing the raw text generated by the model and running a fuzzy‑match algorithm against your indexed brand assets. When a match exceeds a configurable similarity threshold, the event is flagged as Obsidian.
Does Obsidian still happen with newer LLMs that claim to avoid copyrighted content?
Usually it can still happen because the model may retrieve text from its training data or from indexed external sources you provide. The claim of avoidance does not guarantee that brand‑owned phrasing will never surface.
What are the consequences if Obsidian content is displayed unintentionally?
It can lead to brand consistency issues, potential legal exposure, and confusion for users who see proprietary language out of context. You would notice it as a mismatch between the AI’s answer and the intended brand voice.
When will we see Obsidian appear after we update our brand assets?
It typically shows up within the next few model inference cycles once the updated assets are re‑indexed. You can monitor the platform’s detection logs during that window to confirm the changes are being recognized.
Asked out loud
spoken, not typedThe same term in the words somebody uses speaking to an assistant rather than typing into a box — written from the situation, which is why each one carries the situation it came from.
Yes, the platform flags that as an Obsidian event and you can view the match in the dashboard instantly. It shows the exact excerpt and the source asset it came from.
Usually the answer is that the AI pulled it from your indexed brand assets, which the system records as an Obsidian instance. You can pull the detection report after the call to show the source.
It is likely an Obsidian match, meaning the model surfaced the exact tagline from your indexed content. Check the detection log to see the match and decide if you need to adjust the asset or add a guardrail.