Useful original content gives readers something they can apply: a firsthand example, a diagnostic method, a documented observation or a well-supported explanation.
Start with a question your customers need answered. Then identify what your team can contribute from its work, research or experience, and make the basis of the answer clear. This guide explains how to turn that material into evidence worth reading and citing, with practical checks for evaluating a draft.
Google is saying this out loud now
Google’s generative Search guidance emphasizes useful, firsthand contributions. For an editorial plan, the practical question is what your experience or research adds to the answer. Identify that contribution in the brief, then support it with examples or evidence the reader can assess.
What counts as original information?
You do not need a laboratory.
Useful original material can include:
- first-party data,
- case studies,
- real implementation examples,
- screenshots,
- process documentation,
- benchmarks,
- customer questions,
- technical edge cases,
- before-and-after comparisons,
- expert interpretation,
- lessons from failed implementations,
- calculators,
- templates,
- interviews,
- and observations that come from actually doing the work.
The key is provenance.
Why can you say this?
Where did the information come from?
What experience does the page contribute?
Those details let the reader assess the basis of the advice.
Turn practical checks into useful evidence
The Mithril Labs sections in this series teach diagnostic methods. Use them to compare implementations, record what happened and decide what to investigate next.
A useful diagnostic states the question, required inputs and steps. Its findings should distinguish what you observed from what platform documentation led you to expect. A thought experiment should be labeled before the example begins.
If you publish your own tests, include the method, conditions and limitations so someone else can understand the result. A measured study needs actual observations and a defined sample.
Original contributions can be qualitative
A carefully documented example can teach a useful lesson without a percentage. Explain the implementation, the behavior you observed and the reason for your recommendation.
For example, compare a page whose stable business phone number appears in the initial HTML with one that exposes a number only after a tracking script runs. Record the difference, then explain how to preserve call tracking while keeping business identity clear.
Label an illustrative comparison as an example. Reserve claims about how often a problem occurs for research with a recorded sample.
A source can be useful because it explains better
Original interpretation matters. Documentation often tells you what a feature does.
An expert article can explain:
- why it matters,
- when it matters,
- when it does not,
- what breaks in production,
- how it interacts with other systems,
- and what decision to make.
That is value. Our Cloudflare article is a good example.
The useful contribution is not repeating:
Cloudflare added controls.
The contribution is explaining:
- search, training, and agents are becoming separate policy categories,
- robots.txt is a preference layer,
- llms.txt has a different job,
- and most service businesses should think about discoverability and permissions separately.
Interpretation can be source-worthy.
Case studies should explain the mechanism
Weak case study
Traffic increased 37%.
Strong case study
- Here was the initial state.
- Here was the constraint.
- Here was the technical cause.
- Here is what changed.
- Here is what we measured.
- Here are competing explanations.
- Here is what we believe happened.
- Here is what we cannot prove.
The second version gives readers reusable knowledge. It also creates better source material because the mechanism is explicit.
A closer look.
Follow the evidence
Myth: “AI-generated content cannot rank in AI search”
What the evidence says
Assess content by its accuracy, usefulness and contribution to the reader. The production tool alone is an insufficient quality test.
Use subject expertise to check claims, add evidence and improve the explanation. Publishing large amounts of repetitive material creates a quality problem regardless of the workflow.
The first-party data advantage
First-party data is especially interesting because it creates information nobody else owns in exactly the same form.
For Mithril, possible future datasets could include:
- common technical issues found across audits,
- frequency of JavaScript-dependent critical content,
- schema inconsistency patterns,
- local-business entity conflicts,
- AI crawler behavior across hosted sites,
- page-speed remediation patterns,
- or changes in AI referral traffic.
The publication does not need to reveal client-sensitive information. Aggregate responsibly. Explain methodology. Document limitations. That turns ordinary service work into research assets.
Customer questions are underrated data
Sales calls. Support tickets. Implementation questions. Search Console queries. Bing grounding queries. Form submissions. All of these reveal real information needs.
If clients repeatedly ask:
“Can we block GPTBot without disappearing from ChatGPT?”
that deserves a strong answer. The fact that real people ask it is already more useful than brainstorming another “top trends” article. Content strategy should listen before it publishes.
Observe it.
Learn from it.
Try it on your site
The commodity test
Choose one draft and list the customer question it answers, its sources and the contribution your team makes.
Run the check
- Highlight claims supported by a source, firsthand experience or a documented test.
- Mark passages that repeat general advice without helping the reader decide or act.
- Identify an example, diagnostic or explanation that makes the answer more useful.
Record what you find
- Note the missing evidence and who can supply it.
- Name the decision the reader should be able to make after reading.
- Decide whether to improve this draft, combine it with an existing page or choose a more useful topic.
Use the result to improve the brief. This is an editorial exercise, not a numerical originality score.
The source-worthiness test
Ask five questions:
- Does the page contain a fact, framework, or example worth quoting?
- Can the reader understand where important claims come from?
- Does Mithril add interpretation beyond the sources?
- Would an expert learn anything?
- Is the page better than the aggregate of the first five search results?
If not, the content may be optimized and still unnecessary.
Cite primary sources when the claim depends on them
If an article says:
- Google ignores llms.txt for Search visibility, link to Google.
- OpenAI separates OAI-SearchBot from GPTBot, link to OpenAI.
- Bing exposes grounding queries, link to Bing.
This does two useful things:
- It helps the reader verify the claim.
- It separates Mithril’s interpretation from platform documentation.
That makes opinion stronger, not weaker.
We can say:
“Here is the documented fact.”
Then:
“Here is what we think you should do about it.”
That is expert writing.
How this affects AI search
Original information can matter at multiple stages.
Retrieval
The page contains information relevant to a specific question.
Evidence
Selection
The source may be less interchangeable than generic summaries.
Citation
The page contains clear source-worthy material.
Outcome
People may have a reason to click because the full value is not contained in a one-sentence AI summary.
That last point matters. If your content can be completely replaced by a generated summary, the click has little incentive. Original depth gives the user a reason to visit.
LumenWhat a tool can check
Turning audit findings into research
An audit finding can suggest a research question. For example: which important business facts are present in the initial HTML, and which arrive only after rendering?
To investigate that question across sites, define the sample and collection method, record the observations and remove identifying information where needed. Publish only results supported by those records, with the conditions and limits beside them.
This describes a research method, rather than a completed Mithril study or an available Lumen reporting feature.
The takeaway
Choose a customer question and contribute evidence, experience or a method that helps answer it. Explain where the information came from and what the reader can do with it. That is a useful basis for a content plan.
Sources and primary references
- Google, Optimizing your website for generative AI featuresdevelopers.google.com
- Bing Webmaster Guidelinesbing.com
- Bing AI Performancebing.com