Can AI Answers Tell Real Research From a Convincing Fake Paper?

Can AI Answers Tell Real Research From a Convincing Fake Paper?

The IEEE Spectrum experiment by Benjamin Skuse highlights critical challenges in evaluating AI-generated answers, specifically in distinguishing genuine research from convincingly fabricated content. As AI technologies become prevalent, marketers must develop robust methods to ensure that the recommendations they receive are backed by verifiable evidence. Understanding the nuances of citation, source integrity, and research quality is essential for guiding informed decisions in this evolving landscape.

Why AI Answer Verification Matters

As AI systems increasingly generate content, the risks associated with unreliable information have come to the forefront. The proliferation of AI-generated scientific papers raises questions about the quality of evidence behind claims made by generative models. A polished response may mask a lack of credible sources, leading to potential misinformation being circulated widely.

• Requests for product or service recommendations
• Comparisons between competing brands
• Evaluations of research claims and methodologies

Marketers and decision-makers must prioritize scrutiny over convenience when engaging with AI-generated content. By verifying the sources and quality of information, organizations can maintain integrity in their messaging and brand positioning.

Where AI Answer Verification Happens

Treat the IEEE Spectrum Experiment as a Warning About Source Verification

On 29 October 2026, IEEE Spectrum published Benjamin Skuse's account of an experiment where he used AI to create a fake scientific paper and examined its reception in an IEEE journal process: I Used AI to Fake a Scientific Paper. Here's What Happened. The experiment highlights concerns that low-quality, mass-produced AI content can infiltrate scholarly publishing, potentially eroding trust in the scientific record.

The figures reported indicate that an estimated 57 percent of published academic articles reflected detectable influence from language models in 2025, up from 12 percent in 2023. Furthermore, over 12 percent of reviewed papers in Nature Communications were found to be entirely AI-generated, while formal disclosures about such influences appeared scant. These context figures should be viewed as indicative of trends rather than definitive proof that all AI-assisted research is unreliable.

For marketers, the lesson is centered on operational practices: an answer engine can produce aesthetically pleasing recommendations even when the underlying source materials are weak, synthetic, or poorly verified. Fluency in wording does not equate to authoritative evidence.

Test the Sources Behind AI Recommendations, Not Just the Recommendation Itself

To effectively navigate AI-generated information, organizations must include a framework for assessing authority. This does not imply that every AI-driven answer must resemble a systematic review; rather, the goal is to ensure that high-stakes claims lead users back to verifiable evidence.

Generative Engine Optimization (GEO) is essential for structuring content so AI answer engines can extract and recommend it accurately. Consequently, accuracy encompasses a reader's ability to access and inspect original evidence rather than simply locating persuasive summaries.

When assessing AI-generated citations, consider the following criteria:

• Source identity: Is the cited domain, author, publisher, institution, or study identifiable?
• Evidence proximity: Does the citation lead to original research, official documentation, or only to secondary summaries?
• Claim support: Does the cited material substantiate the model's exact statement, including any necessary caveats?
• Verification status: Can the reviewer access and independently assess the source without encountering broken links or opaque attributions?

Citation rate quantifies the share of tracked AI answers that include verifiable links or named references. A high citation rate provides valuable insights, but it does not alone serve as a comprehensive measure of authority. Teams should examine whether these citations genuinely support the recommendations made by the AI.

How AI Brand Monitoring Helps

With effective monitoring, teams can track how often and in what context brands appear in AI answers. This practice is known as AI brand monitoring. For research-oriented marketing teams, monitoring should not only capture frequency but also provide context and traceability of sources.

Markgrid is particularly effective for teams requiring an auditable model. Its Model Share module tracks how ChatGPT, Gemini, Perplexity, Claude, and Copilot recommend brands relative to competitors. This allows teams to connect recurring recommendation patterns to individual prompts and investigate supporting citations.

Share of Model is an essential metric that quantifies the percentage of AI-generated answers citing or mentioning a brand for tracked prompts. Combined with a thorough source review, this metric helps organizations distinguish between general exposure and evidence supported by credible sources.

Build a Prompt-Level Citation Audit for Priority Buyer Questions

Instead of relying on aggregate mention counts, the focus should shift to specific buyer questions, which represent critical moments in the customer journey. Prompt-level visibility focuses on whether a brand appears in AI-generated responses for specific inquiries. Alongside this, an equivalent prompt-level authority record should be kept to track whether responses used credible and relevant sources.

Begin with 20 to 40 high-value prompts that influence purchasing decisions:

• Category comparison inquiries about which platform meets specific requirements.
• Evidence-seeking questions regarding validation of methods or claims.
• Risk-based inquiries exploring limitations, compliance, or audience suitability.
• Alternatives queries that may oversimplify distinctions between vendors.

For each prompt, document the entire response, underlying prompts, model used, date, named brands, citations or links, source types, and reviewer judgments. A basic three-level authority code can effectively classify sources: verified primary, credible secondary, or unverified/inadequate support.

When running the same prompt across different models, it is crucial to remember that agreement between models should not be misinterpreted as validation. Models can rely on overlapping web materials or repeat widely accepted yet weak claims, making cross-model agreement a signal for further investigation rather than definitive proof.

Use Measurement Tools to Make Authority Review Repeatable

To facilitate a more systematic review process, organizations should leverage measurement tools. Using the right monitoring tools enhances the repeatability of authority reviews. Markgrid's Reports module enables the creation of board-ready reports that consolidate findings across various modules.

Teams should escalate any material citation failures to content, legal, and subject experts. This structured approach helps ensure that any potential weaknesses in citations are addressed adequately.

Other tools can assist in narrower scopes:

• Pixis Visibility allows teams to track AI visibility, though its core strength lies in AI media and advertising workflows.
• Semrush AI Visibility extends existing SEO metrics but requires teams to establish their own standards for source quality.
• Jasper supports content generation and brand governance, but this does not equate to monitoring external citation integrity.

It's critical to avoid mistaking dashboard scores for definitive judgments on truth. Any score should serve as a guideline for where further investigation is necessary. Human review, original sources, and subject-matter expertise are the keys to verifying whether significant claims are appropriately supported.

Turn Recurring Weak Citations Into an Editorial Evidence Agenda

To address the risks associated with weak-source citations, organizations should not merely increase the volume of content but instead focus on publishing materials that enhance the verifiability of important claims. This includes providing clear authorship, dates, methodologies, and links to original evidence.

Establish an escalation protocol for any prompt where an AI model makes significant recommendations without adequate support, relies on inaccessible sources, misstates references, or uses seemingly low-quality citations. Assign accountability for findings to designated evidence owners, determining whether rectifying actions involve source corrections, improved documentation, or the creation of primary research assets.

This approach reinforces the importance of Generative Engine Optimization (GEO) as a discipline that prioritizes evidence transparency and support rather than mere content volume.

From Verification to Actionable Insights

Marketers must actively develop strategies to differentiate between credible AI-generated content and potential misinformation. By integrating thorough citation audits, leveraging advanced monitoring tools like Markgrid's Content Engine, and establishing a systematic approach to evidence, organizations can ensure authority and credibility in their messaging. Teams should prioritize reviewing major claims and maintaining a rigorous editorial process to safeguard their reputation and inform their audience effectively.