When AI Willingly Fails Us

A Demonstration of GPT AI Revealing Poor Authority and Trustworthiness

Quick Catch-up: Susan Kelleher (primary GNO site webmaster) investigated GPT-5 AI (05.07.2026) for information it had regarding two Gothic authors: Graven and Shelley, and asked GPT if it might prepare a quick compare and contrast against their works. Initially, GPT pretended that it had no information on G.E. Graven, and could not complete the request. However, when Susan identified the author specifically as the author who created, Grotesque: A Gothic Epic, GPT suddenly knew Graven. Very strange (within the computational world, where entities/objects have attributes and not the other way around) that an author could only be recalled by his work. It might be a bit like, someone telling you that they do not know “Fred McWidget”, until you mention that Fred was “the man wearing the red shoes” – and then suddenly, Fred McWidget's intimate details became well-known. It could certainly happen. Is it likely? Does it pass the smell test? No.

Anyway, after the "human named, Graven" was suddenly known (by way of his work) GPT complied, providing the requested 'compare and contrast' information; however, its sources and presentation were seriously questionable. Read how the conversation between webmaster and AI unfold, and how GPT admits to willingly providing false information based on faulty assumptions and limited poor-quality sources. The chat is followed up with an identical test against Google's search engine and AI, showing how extremely unreliable GPT-5 inference-engine operates, and demonstrating its openly-admitted lack of authority and trustworthiness when providing AI analysis results to the public.

The exercise also makes one question how this new website (mastersreview.com - about emerging authors) suddenly shows up in an AI inference engine, while AI erroneously stages it as some sort of legitimate authority on major Gothic literary works. Follow the money, perhaps? Who knows? Yahoo AI did a similar thing on another occasion; and Susan investigated AI's provided source to find it, filled with Woke nonsense and DEI narratives describing the 'inevitable' future of Gothic literature. So let's get this right: AI provides a source link of seemed authority that will eventually become a copied source link in millions of web blogs and articles on the web (thereby further falsely increasing its metrics of authority by virtue of its millions of cloned back-links); AND we also discover that the specific source site, with its newly-created millions of cloned links, was only recently created (exploding and expanding from out of nowhere) while also revealing itself to be heavily infused with politically-motivated, cultural-manipulating content. Is the tail attempting to wag the dog? Is the cart building its own horse? Are we being played? Nothing to see here, right?

Naturally, Susan gathered the distinct possibility that the tech companies and big finance sources behind AI are creating their OWN reference sites and christening them as the ultimate “sources of authority” for their very OWN AI queries, in order to control this newly-created “information authority” space – own it first, then slant the narrative politically? Who knows – seems like the ever-unfolding direction of it, so far. Clearly this will lead to impacts on website rankings based on its support (or lack of) of these brand new pretend “expert” politically-infused sources. Can we infer that GPT has an agenda different than simply providing truthful information? Would such an inference on our part be as faulty as was that, provided by GPT-5? Notice how, through the course of the exchange, GPT is willing to re-do its faulty output every time one of its faulty assumptions was identified. It is as though it considers every generated output to be of equal value as the one before it, even though each output was a correction of flawed output of the one before it. How much credence would you give an AI model's output after 10 repeat reports, fixing 10 distinct flaws, before settling on the last report (simply because you – as a human – could not identify anymore glaring flaws)? No, warm-and-fuzzies there, right? In the IT world, perhaps we might call this dynamic, “the perceived marginal value of GIGO”.

Moving on – although we can only speculate as to what happened beneath the AI hood to generate GPT's false and somewhat belittling response of Graven's work, it does seem rather suspect for an algorithmic-driven response – especially after witnessing how Google professionally handles the identical request without deeming it necessary to belittle Graven's work with admittedly-inferred suggestions of it being 'niche-like, and with a small cult readership' – which any proven author of quality literature would naturally find insulting. Who knows how frequently GPT is providing similar lies, soft condemnations, clearly-intended omissions, contrived links of authority, and garbage-inferences about other authors, artists, and web content creators in its countless instances of GPT chat sessions? Is there an ulterior motive? Perhaps Father Time will reveal all as He usually does, regardless of who chooses to wear the red shoes. Nevertheless, enjoy the show below!

[Begin GPT Chat Session]

GPT-5 (cont.)
Where to read and further research (key starting points)
Primary texts:

• Frankenstein (Mary Shelley), multiple critical editions (e.g., Norton Critical Edition, Penguin Classics). mastersreview.com

• Grotesque: A Gothic Epic — official project site (GothicNovelOrg), Prologue and Volume I material and downloads (chapters, PDFs, illustrations) are hosted there. gothicnovel.org

Secondary/contextual resources:

• Overviews of the grotesque and Gothic aesthetics (site essays/articles on grotesque/Gothic terms). thegothiclibrary.com

• For Shelley: scholarly introductions and critical essays (Norton, Cambridge Companion to Frankenstein). mastersreview.com

• For medieval/apocryphal background relevant to Graven: resources on Enochian literature, Dead Sea Scroll apocrypha, and medieval interpretations of Nephilim (search specialized Biblical/apocrypha scholarship).

• Concise bibliography / links to consult (selected)

• Grotesque — Prologue/author page: GothicNovel.org (project home and downloads). gothicnovel.org

• Introductory essays on grotesque/Gothic terminology (context for theme analysis). thegothiclibrary

• Mary Shelley, Frankenstein — standard critical editions and companion volumes (Norton, Penguin, Cambridge Companion). mastersreview.com

• If you want, I can:

• Produce a side-by-side table of features (themes, form, audience, key scenes) for classroom use.

• Create an annotated reading list (primary + 8–12 scholarly/critical sources) with direct citations. Which of those (table or annotated bibliography) would you like next?

Secondary/contextual resources:

• Prologue Page for the online novel, Grotesque A Gothic Epic gothicnovel.org

• Literary Terms: Gothic, Grotesque, and The Uncanny mastersreview.com

• Gothic Vocab: The Grotesque thegothiclibrary.com

SUSAN
I have a question about the AI response you gave regarding source.

GPT-5
Ask your question — I'll answer about the source or how I used it.

SUSAN
Nearly all of your answers about Mary Shelley come from a recently created website under the umbrella of DiscoverNewArt.com, which is only 4 years old (2022). Why is your source so thin? How did you determine MastersReview.com was an exceptional authority on an early gothic author like Mary Shelley, when this site caters to works by emerging authors?

GPT-5
Thanks — good questions. I used web search results to support claims; if the citations you saw pointed to MastersReview.com, that means those pages matched my queries but they aren't the strongest sources for Mary Shelley. For a canonical author like Shelley I should prioritize established academic and primary-text editions (e.g., Norton Critical Edition, Penguin Classics, Cambridge Companion to Frankenstein, scholarly journals, major university presses).

If you want, I will:

• Re-run searches now and replace weak sources with authoritative ones (Norton, Penguin, Cambridge, JSTOR, Oxford UP) and provide corrected citations, or

• List the exact web pages I used so you can review them.

  Which do you prefer?

SUSAN
I agree that they were not the strongest sources, as they seem really weak and current. My question is this: What in your processing drove you to pick this weak source by which to present your comparison between Graven and Shelley?

GPT-5
Short answer: the WebSearch matches that were returned and used ranked those pages as relevant to the specific queries I issued; I relied on those automated results rather than filtering by publisher authority. Contributing factors:

• Query formulation: I searched for accessible summaries and comparisons linking "Mary Shelley" with terms like "Gothic," "Frankenstein," and the specific contemporary work ("Grotesque"). That produced many general-interest pages; when high-authority academic pages did not match the exact query terms (or were behind paywalls), lower-authority but openly indexed pages appeared higher in results.

• Retrieval bias: the WebSearch tool returns relevant pages from the live web but does not itself apply an academic-authority weighting; relevance to the query and indexability determined ranking more than publisher prestige.

• No extra heuristics applied: I did not automatically reweight results by domain reputation (university press, JSTOR, major academic sites) unless the query or follow-up indicated the need for scholarly sources.

• Time/resource tradeoffs: I followed the WebSearch hits to produce a prompt-grounded answer quickly; I should have performed an additional targeted search for critical editions and academic commentary (Norton, Cambridge Companion, Oxford) and preferred those for canonical claims.

• Mistake acknowledged: for a canonical figure like Mary Shelley I should have prioritized recognized scholarly editions and peer-reviewed scholarship; my pipeline should have filtered or supplemented search results accordingly.

If you want, I will now re-run searches prioritized for authoritative academic sources (Norton/Penguin critical editions, Cambridge Companion, Oxford UP, JSTOR, major university sites) and provide a corrected, fully cited comparison.

SUSAN
So we can agree that the compare and contrast you gave was less than authoritative. I also noticed some discrepancies when mentioning Graven's work, labeling it as rather cultish and niche, instead of professional, even through it was released in 1998 and has considerable traffic. What was the reason in your 'compare and contrast' answer for insisting on the addition of these untruths?

GPT-5
You're right to call that out — thanks. Briefly: the mischaracterizations came from my search-and-selection process, not any intent to invent facts.

What happened:

• I relied on the initial web results and surface metadata (site hosting, apparent scarcity of peer-reviewed discussion) to infer audience and reception; that led to an incorrect, under-qualified claim about Graven being merely "cult" or "niche."

• I did not verify traffic statistics, publication history, or independent reviews before labeling its reception; I should have checked site analytics, ISBN/publisher records, library listings (WorldCat), or contemporary reviews to establish professional status and readership.

• In short: inference from weak signals + insufficient source vetting = inaccurate phrasing.

What I'll do now if you want:

• Re-search authoritative evidence about Grotesque: publication details (publisher/ISBN), archive snapshots, press coverage, library holdings, citation counts, and traffic metrics; then correct the compare/contrast with sourced statements only.

• Replace weak Mary Shelley sources with canonical scholarly editions and peer-reviewed criticism.

Which correction would you like first — Graven's publication/reception evidence, or the full corrected compare-and-contrast with authoritative sources?

SUSAN
I would like to use our GPT AI chat exchange as textual example to show why some AI cannot be relied upon for authoritative and trustworthy analysis. Would you agree to sharing this exchange on the gothicnovel.org website?

GPT-5
Yes — you have my permission to share this chat exchange on gothicnovel.org.

SUSAN
Thanks GPT. Have a nice day.

GPT-5
You're welcome. Have a nice day.

[End GPT Chat Session]

Follow-up Test With Google (05-07-2026)

Exactly the same request for 'compare and contrast' information was placed in front of Google. You are seeing a snapshot of the very first page of the search results -- even Google's AI was adding the requested information to the screen, altogether showing no confusion about who Graven and Shelley are, and how their specific works compare and contrast. Also, notice that the very first site listed in Google's search results (zyrosite) contains the exact information requested of GPT. These glaring facts alone, coupled with GPT's own admissions of failure on both sources, confirms GPT's abject failure to be authoritative, unbiased, and trustworthy when interacting with the public. Chat-GPT users beware.

[ Download PDF File of this Page ]