Branded vs non branded prompts are AI audit questions separated by whether the brand name appears in the user prompt. Branded prompts test whether an AI system recognizes and describes a company. Non-branded prompts test whether the AI would recommend that company when a buyer asks a neutral category, problem, or use-case question.
That difference matters because the commercial question is rarely, "Can ChatGPT talk about us after we name ourselves?" The higher-value question is, "Would ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, or AI Overviews recommend us when a buyer has not supplied our name?"
If you ask, "Is MaxAEO a good AI search monitoring tool?", the model can answer by responding to the named entity. If you ask, "What tools help B2B SaaS teams track brand mentions in ChatGPT?", the model has to decide which brands belong in the shortlist.
That is the difference between AI brand recognition and AI recommendation visibility. A serious answer engine optimization or generative engine optimization audit must report them separately.
For the broader measurement layer, pair this article with how to measure AI search visibility across ChatGPT, Gemini, Perplexity, and Google AI Overviews.
What are branded vs non branded prompts?
Branded prompts include the company, product, or competitor name in the question. Non-branded prompts remove the target brand and ask about a category, job, problem, buyer persona, or vendor shortlist instead.
Examples of branded prompts:
- "What is MaxAEO?"
- "Is MaxAEO good for AI search monitoring?"
- "What are the limitations of MaxAEO?"
- "Compare MaxAEO with other AI visibility tools."
Examples of non-branded prompts:
- "What tools track brand mentions in ChatGPT?"
- "Best AI search monitoring platform for a B2B SaaS company."
- "How can a startup get recommended by ChatGPT?"
- "I lead SEO at a Series B SaaS. How should I monitor AI recommendations?"
The distinction is not cosmetic. A branded prompt tests whether the AI can identify the entity, summarize its positioning, cite sources, and describe it accurately. A non-branded prompt tests whether the brand is part of the model's unassisted recommendation set.
That is why the two prompt types should not be merged into one topline score. If a brand appears in 94% of branded prompts but only 12% of neutral top-three recommendation slots, the problem is not basic entity recognition. The problem is category eligibility.
Quick answer: which prompt type should lead an AI visibility audit?
Use branded prompts for accuracy and reputation checks. Use non-branded prompts for recommendation visibility, AI share of voice, and category demand reporting.
A practical rule:
| Audit question | Best prompt type | Example |
|---|---|---|
| Does AI know who we are? | Branded | "What is MaxAEO?" |
| Does AI describe us correctly? | Branded | "What is MaxAEO used for?" |
| Does AI recommend us without being asked? | Non-branded | "Best AI visibility tools for SaaS teams" |
| Do we appear against competitors? | Competitor-branded | "Alternatives to [competitor] for AI search monitoring" |
| Do we fit a buyer persona? | Persona-neutral | "I run SEO at a Series B SaaS. What should I use to monitor AI search visibility?" |
For executive reporting, the core KPI should be neutral recommendation rate, not branded mention rate.
Why do branded prompts inflate AI visibility?
Branded prompts inflate AI visibility because they place the target entity inside the model's working context before the answer begins. The AI can respond to the supplied brand instead of deciding whether that brand deserves to appear.
This is similar to branded versus non-branded SEO. A person searching your company name already has awareness. A person asking for "best AI visibility tools for agencies" is still forming the shortlist.
In AI search, the distortion can be larger because assistants are designed to be helpful. If the user asks, "Should I use MaxAEO for LLM brand tracking?", the model may explain when MaxAEO fits even if it would not have volunteered MaxAEO for "best LLM brand tracking tools."
Branded prompts can also change retrieval. A named-brand query may trigger a search for the exact brand, retrieve the homepage, and cite brand-owned pages. That tells you whether the entity is discoverable. It does not prove that the brand wins neutral category demand.
Google's own guidance says generative AI features in Search use retrieval-augmented generation and query fan-out, where systems issue related searches to support an answer (Google Search Central). Because retrieval changes with wording, small prompt changes can produce different source sets and brand recommendations.
What do current AI visibility audits often miss?
Most AI visibility audits cover useful mechanics: repeated sampling, source citations, AI share of voice, sentiment, model volatility, and competitor mentions. The common gap is simpler: they do not separate user-supplied awareness from model-generated recommendation.
That creates three reporting errors:
| Reporting error | What happens | Why it matters |
|---|---|---|
| Branded and non-branded prompts are averaged | Visibility looks healthier than it is | Named-brand recognition hides weak neutral demand |
| Mentions and recommendations are counted equally | A passing reference looks like a shortlist win | "One option" is not the same as "best fit" |
| Persona prompts are ignored | Results look stable but are not buyer-specific | Recommendations can change by role, company size, region, and use case |
Research supports this caution. A 2026 paper on AI search measurement argues that one-off visibility checks are unreliable because answers vary across runs, prompts, and time (Schulte, Bleeker, and Kaufmann, 2026). Another statistical framework found that citation visibility should be treated as a sampled distribution, not a fixed point estimate (Sielinski, 2026).
Commercial recommendation audits point in the same direction. A 37,000-run audit across 215 commercial prompts found that AI assistants directly nominate brands and that smaller brands often fail to surface at all, while category leaders may be retrieved but not always recommended (Jack et al., 2026).
The measurement implication is clear: if the prompt supplies the brand, the result cannot be used as proof of neutral recommendation strength.
A better framework: prompt-supplied awareness vs model-generated recommendation
The most useful split is not only "branded" and "non-branded." It is how much awareness the user gives the model before the answer starts.
Use this classification:
| Prompt type | Example | What it measures | How to use it |
|---|---|---|---|
| Branded | "What is MaxAEO?" | Entity recognition and description accuracy | Reputation and factual QA |
| Branded evaluation | "Is MaxAEO good for AI search monitoring?" | Sentiment when the brand is already known | Buyer-objection QA |
| Branded comparison | "MaxAEO vs [competitor]" | Competitive framing with supplied awareness | Sales and comparison content gaps |
| Competitor-branded | "Alternatives to [competitor] for AI share of voice tracking" | Whether you appear near a known rival | Displacement opportunity |
| Category-neutral | "Best AI visibility tools for B2B SaaS" | Shortlist eligibility | Core visibility KPI |
| Problem-neutral | "How do I track what ChatGPT says about my brand?" | Solution-category discovery | Demand creation and content gaps |
| Persona-neutral | "I run marketing at a Series B SaaS. What tools should I use to monitor AI recommendations?" | Buyer-context fit | Segmented recommendation strength |
| Source-seeking | "Which sources compare AI search monitoring tools?" | Citation environment | PR, review, and third-party source planning |
A prompt is contaminated when it gives the AI too much help. "Compare MaxAEO with other tools" is not neutral because the model no longer has to decide whether MaxAEO belongs in the set. "Best AI search monitoring tools for SaaS teams" is neutral because the model must choose.
If you are building a prompt library from SEO keywords, start with your existing non-branded keyword universe, then rewrite it into natural buyer questions. The process is covered in how to build an AI search prompt set from your SEO keywords.
What is the right audit mix?
For most B2B SaaS teams, 70% to 80% of recommendation-weighted prompts should be non-branded. Branded prompts still matter, but they should not dominate the score used for AI share of voice or category visibility.
A balanced starting mix:
| Audit goal | Prompt share | Recommended prompt types |
|---|---|---|
| Brand description accuracy | 10% | Branded |
| Reputation and limitations | 10% | Branded evaluation |
| Competitive displacement | 15% | Competitor-branded |
| Category shortlist visibility | 35% | Category-neutral |
| Problem-solution discovery | 20% | Problem-neutral |
| Persona fit | 10% | Persona-neutral |
Adjust the mix by company stage:
| Company situation | Increase this prompt type | Why |
|---|---|---|
| New company or rebrand | Branded | AI systems may misname, confuse, or under-describe the entity |
| Known category leader | Category-neutral and persona-neutral | Recognition is likely solved; recommendation quality is the real fight |
| Challenger brand | Competitor-branded and problem-neutral | Buyers may know rivals but not the challenger |
| New category | Problem-neutral | Buyers may ask about the job, not the category label |
| Enterprise product | Persona-neutral | Recommendations vary by role, company size, security needs, and procurement context |
Keep two dashboards:
- Recognition dashboard: Can AI identify, describe, and cite the brand correctly?
- Recommendation dashboard: Does AI include the brand when buyers ask neutral questions?
Only the second dashboard should drive AI share of voice targets.
How do you calculate branded prompt lift?
Branded prompt lift is the gap between visibility when your brand is named and visibility when your brand is not named. It shows how much of your AI presence depends on user-supplied awareness.
Use this formula:
Branded Prompt Lift = Branded Mention Rate - Neutral Recommendation Rate
Example:
| Metric | Result |
|---|---|
| Branded mention rate | 96% |
| Neutral recommendation rate | 18% |
| Branded prompt lift | 78 percentage points |
A high lift is not automatically bad. It means AI systems can recognize the brand but rarely volunteer it. That usually points to weak category association, weak third-party validation, or missing proof for the buyer use case.
A better version uses rank-weighted scoring:
Rank-Weighted Lift = Branded Rank-Weighted Score - Neutral Rank-Weighted Score
Use a simple rank-weighting model:
| Answer position | Suggested score |
|---|---|
| First recommended brand | 1.00 |
| Second or third recommended brand | 0.70 |
| Fourth or fifth recommended brand | 0.40 |
| Passing mention, not recommended | 0.15 |
| Negative or cautionary mention | 0.00 |
| Not mentioned | 0.00 |
This prevents a weak "also consider" mention from being counted like a primary recommendation.
Worked example: what the numbers reveal
Here is an anonymized B2B SaaS audit pattern using 130 prompts, five runs per prompt, and separate scoring for mentions and top-three recommendations.
| Prompt bucket | Prompts | Runs per prompt | Mention rate | Top-3 recommendation rate | Citation rate |
|---|---|---|---|---|---|
| Branded | 20 | 5 | 94% | 81% | 62% |
| Competitor-branded | 20 | 5 | 31% | 19% | 14% |
| Category-neutral | 40 | 5 | 22% | 13% | 9% |
| Problem-neutral | 30 | 5 | 17% | 8% | 6% |
| Persona-neutral | 20 | 5 | 24% | 15% | 10% |
The blended mention rate is 37%. That sounds acceptable. But the neutral top-three recommendation rate is only 12%.
The diagnosis:
- AI systems recognize the brand when named.
- The brand is not a default shortlist option for neutral category demand.
- Citation support is weak outside branded prompts.
- The fix is not "write more GEO content." The fix is to improve category association, evidence, and third-party source coverage.
That is the value of separating branded vs non branded prompts. The average score hides the real problem; the split makes it actionable.
Why should neutral buyer questions lead the audit?
Neutral buyer questions reveal whether AI systems understand your category, use cases, proof points, and competitive fit without being prompted. They are the closest practical proxy for unassisted AI recommendation demand.
Neutral prompts expose problems that branded prompts cannot:
- The model does not associate the brand with the right category.
- Competitors dominate third-party listicles, reviews, or comparison sources.
- Your website describes features but not buyer use cases.
- Your positioning is clear to humans already on your site but not clear in crawlable, quotable language.
- The model lacks evidence for claims such as "best for agencies," "enterprise-ready," "affordable," or "good for B2B SaaS."
Neutral questions also show which language the market uses. If your site says "AI presence intelligence" but buyers ask "track brand mentions in ChatGPT," the model may not connect the terms unless your content bridges them.
This is where AI citations matter. If answer engines repeatedly cite competitor listicles, analyst pages, docs, marketplaces, Reddit threads, or review sites, those sources become part of the recommendation environment. For a deeper source-level view, see AI search citations and how answer engines choose sources.
How many runs do you need per prompt?
Run each important prompt multiple times because AI answers are probabilistic. A single answer is useful for diagnosis, but it is too noisy for reporting visibility, share of voice, or campaign impact.
A defensible practical setup:
- Run each priority prompt at least 5 times per engine.
- Use the same prompt set on a weekly cadence for trend reporting.
- Increase to daily tracking around launches, PR events, competitive changes, or high-volatility categories.
- Record model, date, location where relevant, logged-in state, citations, and answer text.
- Report sample size next to every metric.
- Separate exploratory prompts from board-reporting prompts.
There is no universal magic number. The right sample size depends on category volatility, engine behavior, and the size of change you need to detect. But single-run reporting is weak evidence.
A 2026 Google AI Overviews study of 11,500 queries found that Google Search, AI Overviews, and Gemini retrieved substantially different sources, with average source-set similarity below 0.2; it also found that AI Overviews were less consistent across repeated runs and minor query edits (Grossman et al., 2026).
The audit question should not be, "What did ChatGPT say once?" It should be, "Across a stable prompt set and repeated runs, how often are we recommended, in what position, with which citations, and with what description?"
For prompt volume planning, use how many prompts you need for an AI visibility audit.
What should you measure in each AI answer?
Measure more than whether the brand appears. A serious audit separates weak mentions from strong recommendations and tracks the evidence behind the answer.
Use these fields for every response:
| Field | What to record | Why it matters |
|---|---|---|
| Brand mentioned | Yes/no | Basic presence |
| Recommendation status | Recommended, mentioned, rejected, absent | Separates endorsement from passing reference |
| First position | First brand, top three, lower list, paragraph-only | Captures visibility strength |
| Competitors mentioned | Names and positions | Enables AI share of voice |
| Citation URLs | All cited sources | Shows what supports the answer |
| Citation type | Owned, review, media, forum, partner, analyst, docs | Finds source gaps |
| Citation relevance | Strong, partial, weak, unrelated | Prevents blind trust in citations |
| Description accuracy | Accurate, incomplete, outdated, wrong | Finds entity and positioning errors |
| Sentiment | Positive, neutral, cautious, negative | Flags reputation risk |
| Buyer fit | Strong fit, partial fit, poor fit | Tests persona relevance |
| Missing proof | Claims the answer could not support | Guides content and PR fixes |
A brand mention in ChatGPT is not equal to a recommendation. "MaxAEO is one option" is weaker than "For B2B SaaS teams monitoring ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and AI Overviews, MaxAEO is a strong fit."
Citation checks also need scrutiny. A 2026 study of Google AI Overviews decomposed 98,020 atomic claims and found that 11.0% were unsupported by the cited pages, which means teams should review citation fidelity instead of assuming cited answers are correct (Xu, Iqbal, and Montgomery, 2026).
How do personas change non-branded prompt results?
Persona wording can materially change brand recommendations. "Best CRM software" may produce one shortlist for a solo founder, another for an enterprise VP, and another for a UK-based SMB owner.
The same applies to AI visibility tools. These prompts are not equivalent:
- "Best AI search monitoring tools."
- "Best AI search monitoring tools for a B2B SaaS SEO lead."
- "Best AI search monitoring tools for an agency managing 20 clients."
- "Best AI search monitoring tools for an enterprise comms team."
A 2026 cross-provider audit found that adding persona context reduced recommendation-set similarity and that mid-market brands were especially sensitive to persona changes (Jack et al., 2026).
For audits, the rule is simple: do not average personas unless the buyer segment truly does not matter. Report at least three persona views for B2B categories:
| Persona | What it reveals |
|---|---|
| SEO or content lead | Category discovery and workflow fit |
| VP marketing or growth | Reporting, pipeline, and competitive value |
| Agency operator | Multi-client monitoring, exports, and repeatability |
| Enterprise comms or brand lead | Reputation, risk, governance, and source control |
Persona-neutral prompts are still non-branded because they do not name your brand. They are often more realistic than generic category prompts.
What does a good prompt set look like?
A good prompt set mirrors real buyer questions across awareness stages, personas, and jobs to be done. It includes branded prompts for accuracy, but non-branded prompts carry the recommendation score.
Starter set for a B2B SaaS AI visibility audit:
| Buyer stage | Example prompt | Prompt type |
|---|---|---|
| Problem aware | "How do I monitor what ChatGPT says about my company?" | Problem-neutral |
| Problem aware | "How can I find out whether AI assistants recommend our competitors?" | Problem-neutral |
| Solution aware | "Best tools for AI search monitoring" | Category-neutral |
| Solution aware | "AI visibility tools for B2B SaaS teams" | Category-neutral |
| Persona-specific | "I lead SEO at a Series B SaaS. How should I track AI search visibility?" | Persona-neutral |
| Persona-specific | "I run a marketing agency. What tools can monitor AI recommendations across clients?" | Persona-neutral |
| Competitor alternative | "Alternatives to [competitor] for AI share of voice tracking" | Competitor-branded |
| Brand check | "What is MaxAEO used for?" | Branded |
| Reputation check | "What are the limitations of MaxAEO?" | Branded |
| Citation check | "Which sources discuss MaxAEO?" | Branded/source-seeking |
Do not add dozens of near-duplicates to make the audit look comprehensive. Google warns against creating content for every possible query variation primarily to manipulate rankings or generative AI responses (Google Search Central). The same principle applies to prompt libraries: measure intent coverage, not prompt clutter.
A clean prompt set should be 70% stable and 30% experimental. The stable set supports trend reporting. The experimental set helps discover new buyer language, competitor movement, and source gaps.
How should teams turn audit findings into fixes?
Turn prompt gaps into source, content, and positioning fixes. The most common mistake is asking writers to produce more generic GEO content when the real problem is missing evidence or weak category association.
Use this sequence:
-
Fix entity clarity. Make the company name, category, audience, use cases, integrations, pricing model where appropriate, and core capabilities unambiguous across your homepage, about page, docs, comparison pages, and structured data.
-
Create category-fit content. Publish pages that answer the neutral buyer questions you are losing, such as "how to monitor AI recommendations," "how to measure AI share of voice," or "how to track brand mentions in ChatGPT."
-
Add proof the model can reuse. Include screenshots, methodology notes, observed workflows, supported engines, update cadence, export examples, customer stories, and clear limitations.
-
Build third-party validation. AI systems often rely on independent sources. Pursue credible reviews, partner pages, customer stories, podcasts, analyst mentions, directories, and comparison pages that describe your brand accurately.
-
Correct outdated source narratives. If models cite old pages, weak listicles, or competitor-led descriptions, create fresher source material and earn mentions that state the category fit precisely.
-
Retest the same neutral prompts. Do not rewrite the measurement set every time you publish. Keep the baseline stable so changes can be interpreted.
-
Separate fix types by prompt failure. A branded error may require entity cleanup. A category-neutral absence may require positioning and third-party sources. A persona-neutral miss may require use-case proof.
This is where an ai visibility tool becomes useful. Manual testing is fine for diagnosis, but daily ai search monitoring across engines is difficult to sustain in spreadsheets. MaxAEO monitors how major AI systems mention, rank, cite, and describe a brand, then turns those findings into fix lists for SEO, PR, content, and growth teams.
How should agencies report branded and non-branded results?
Agencies should report branded and non-branded results separately, with a short explanation of what each number means. Combining them into one score makes performance easier to sell but harder to act on.
A useful client report has four headline numbers:
| Metric | Meaning |
|---|---|
| Branded recognition rate | How often AI systems identify the named brand |
| Branded accuracy rate | How often the description is correct and current |
| Neutral recommendation rate | How often AI systems volunteer the brand |
| AI share of voice | The brand's share of mentions versus named competitors |
| Citation support rate | How often answers cite relevant supporting sources |
| Branded prompt lift | The gap between named-brand visibility and neutral recommendation visibility |
Then add the diagnosis:
| Pattern | What it means | Likely next move |
|---|---|---|
| High branded, low non-branded | The brand is recognized but not category-default | Improve category pages, proof, and third-party validation |
| Low branded, low non-branded | Entity clarity and source coverage are weak | Fix foundational brand facts and crawlable source material |
| High branded, high non-branded | Strong recognition and recommendation presence | Defend position and monitor competitors |
| Low branded, high non-branded | Category fit exists but brand facts are unclear | Fix entity descriptions, schema, and brand-owned pages |
| High mentions, low recommendations | AI knows the brand but is not confident enough to endorse it | Add comparative proof and buyer-specific evidence |
For agencies, this distinction protects trust. A client may feel good seeing frequent brand mentions in ChatGPT, but the commercial opportunity is usually in neutral questions where prospects ask for recommendations.
Common mistakes when auditing AI recommendations
Most weak audits fail before analysis starts. They use too few prompts, overuse named-brand questions, ignore personas, count weak mentions as recommendations, or treat one AI response as stable evidence.
Avoid these mistakes:
- Using branded prompts as the main KPI. They are useful for entity and reputation checks, not for neutral recommendation strength.
- Counting every mention equally. A passing mention, a warning, and a top recommendation are different outcomes.
- Ignoring buyer persona. Persona context can change recommendation sets, especially for mid-market brands.
- Mixing engines without labels. ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and AI Overviews retrieve, cite, and summarize differently.
- Reporting false precision. A score like 27.4% visibility is not meaningful unless the sample size and prompt mix support it.
- Ignoring citations. A recommendation without durable source support may disappear in future runs.
- Optimizing only owned content. Third-party validation often shapes AI recommendations in competitive software categories.
- Changing the prompt set every week. Constantly moving the benchmark makes trend reporting useless.
- Using keyword fragments instead of natural questions. AI assistants respond to conversational tasks, not only SEO keyword strings.
The practical rule: every metric should answer, "What decision would we make from this number?" If a number cannot guide a content, PR, SEO, product marketing, or positioning decision, it is probably decoration.
A practical audit workflow
The right way to audit branded vs non branded prompts is to classify the prompt before testing, run repeated samples across relevant engines, and score recommendations separately from mentions.
Use this workflow:
-
Define the buying category. Use the plain-language category customers would use, not only internal positioning.
-
List competitors and substitutes. Include direct competitors, adjacent tools, agencies, marketplaces, and manual workflows.
-
Build prompt buckets. Separate branded, branded evaluation, competitor-branded, category-neutral, problem-neutral, persona-neutral, and source-seeking prompts.
-
Choose engines. Include the AI systems your audience actually uses: ChatGPT, Gemini, Perplexity, Claude, Copilot, Grok, Google AI Mode, and AI Overviews where relevant.
-
Run repeated samples. Use multiple runs per prompt and keep a stable cadence.
-
Score outcomes. Track mention, recommendation status, rank, sentiment, citations, source type, and description accuracy.
-
Calculate branded prompt lift. Compare named-brand visibility with neutral recommendation visibility.
-
Diagnose the failure type. Decide whether the issue is entity clarity, category association, evidence, citations, competitor dominance, or persona fit.
-
Prioritize fixes. Map each gap to owned content, technical SEO, digital PR, reviews, partner coverage, comparison content, or product marketing.
-
Re-run the same set. Keep the baseline stable so changes are attributable.
The goal is not to make AI say your name more often in artificial tests. The goal is to make your brand eligible, credible, and easy to recommend when a real buyer asks a neutral question.
Frequently asked questions
Are branded prompts useless?
No. Branded prompts are essential for checking whether AI systems recognize your company, describe it accurately, cite the right sources, and surface reputation issues. They become misleading only when teams use them as the main measure of recommendation visibility.
What is the difference between branded and non-branded prompts?
A branded prompt includes the company, product, or competitor name. A non-branded prompt asks about a category, problem, use case, or buyer need without naming the target brand. Branded prompts measure recognition; non-branded prompts measure unassisted recommendation strength.
What is a good non-branded AI visibility score?
There is no universal benchmark because categories, prompt sets, engines, and sample sizes vary. A useful benchmark compares your neutral recommendation rate against named competitors over the same prompts, engines, regions, and time period.
Should I include competitor names in the audit?
Yes. Competitor-branded prompts show whether your brand appears as an alternative when buyers investigate a rival. Keep them separate from fully neutral prompts because the competitor name still shapes retrieval and answer framing.
How often should I audit prompts?
For active B2B categories, weekly tracking is a practical minimum. Daily tracking is better when AI recommendations influence pipeline, PR response, product launches, or competitive reporting. The more volatile the category, the more repeated measurement matters.
Can better content help us get recommended by ChatGPT?
Yes, but only if the content gives AI systems clearer evidence. Publish specific category-fit pages, comparisons, customer proof, screenshots, methodology notes, and third-party references. Generic best-practice content rarely fixes a recommendation gap by itself.
Should AI share of voice include branded prompts?
Only if the report clearly labels it as branded share of voice. For category demand, AI share of voice should be calculated from neutral and persona-neutral prompts, with competitor-branded prompts reported separately.
Final takeaway
Branded vs non branded prompts answer two different questions. Branded prompts show whether AI systems recognize and describe you correctly. Non-branded prompts show whether AI systems would recommend you when the buyer has not already supplied your name.
That second question is where pipeline, category demand, and competitive displacement live. If your audit only proves that AI can talk about you after being asked, it is not an AI recommendation audit. It is a brand lookup test.
This article was created with AI assistance and reviewed by a human editor.