The market now offers everything from £20 “rank first on ChatGPT” gigs to five-figure programmes with citation-multiplier guarantees. We sell GEO ourselves, so we have every commercial incentive to say yes. We read the research instead. Verdict up front: no · nobody can guarantee what an AI will say, including us. Confidence: strong. Here is the evidence, openly weighed.
chance two runs of the same prompt return the same brand list · SparkToro’s 2,961-prompt reproducibility study with 600 volunteers (2025)
of the domains AI answers cite change within a month · published by Profound, an AI-visibility platform, from its own tracking data (2026)
of 62,100+ AI-crawler visits touched the llms.txt file in Otterly.AI’s 90-day experiment · an honestly published null result (Feb 2026)
Google’s official May 2026 guidance: no special files, no AI-specific rewriting needed for its generative features · the standard fundamentals are the lever
The reproducibility evidence is the foundation: when SparkToro ran 2,961 prompts with 600 volunteers, the chance of the same prompt producing the same brand list twice was under one in a hundred. Rand Fishkin's conclusion: “Any tool that gives you a ‘ranking position in AI’ is full of baloney.” There is no stable ranking to guarantee a position in.
The platforms selling measurement agree, in their own data. Profound, the category's largest platform, publishes that 40 to 60% of the domains AI answers cite change within a month, and that platform drift, not sampling, is the limit of measurement precision. Notably, none of the twelve enterprise AI-visibility platforms we reviewed carries a citation guarantee: the companies with the most measurement data will not promise the outcome. The guarantees live at the opposite end of the market, where the evidence is thinnest and the terms and conditions are thickest.
The engine operators say the quiet part officially: Google's May 2026 guidance states that its generative features need no special files or AI-specific tricks, just the fundamentals. And the honestly published null results (Otterly's 90-day llms.txt experiment: 0.1% of AI-bot visits) show how thin some heavily marketed tactics are. What actually correlates with being cited, in the research and in our own measurement work: readable structure, verified entity signals, original quotable data, and genuine third-party authority. Correlates. Not guarantees.
Our position, stated with its incentive on the table: Mode sells GEO at fixed published prices, so honest limits cost us sales to anyone who prefers a promise. We publish this anyway, along with our own indexes and our own misses, because the alternative is joining a class of claims that UK advertising rules (CAP Code 3.7) exist to catch. What would change our mind is on record in the questions below.
No. AI answers are probabilistic: SparkToro’s 2,961-prompt study found under a 1% chance that the same prompt returns the same list of brands twice, and platform-published data shows 40 to 60% of cited domains change within a month. Nobody controls the output of an engine they do not run. Providers offering guarantees are offering refund mechanics defined in their own terms, not control over AI answers.
Read the terms. The pattern across the market runs from £20 gig-site promises to “rank 1st on ChatGPT”, through refund-if-not-featured offers, to citation-multiplier guarantees hedged by eligibility conditions and minimum terms. In each case the guarantee is a marketing device whose escape hatch is the fine print. Under UK advertising rules (CAP Code 3.7, substantiation), objective claims need evidence, and the evidence base here points the other way.
No, it means GEO is probabilistic. The measurable levers are real: machine-readable structure, verified entity signals, quotable original content, and third-party authority correlate with being retrieved and cited, and the research (including the 2023 Princeton GEO study, properly read: up to ~40% relative visibility lift in a controlled benchmark, not a guarantee on any live brand) supports improving them. The honest offer is exactly that: improve the signals, measure the movement, publish the misses along with the hits.
If an AI engine sold placement with a service-level agreement, citations would become buyable and guaranteeable, as advertising. Early paid AI-answer formats are emerging, which rather proves the point: the organic side cannot be guaranteed, which is why the engines themselves are building the paid side. We will update this page when the evidence changes; it is date-stamped for that reason.
Sources: SparkToro prompt-reproducibility study (2,961 prompts, 600 participants, 2025) · Profound published methodology and drift statistics (2026) · Otterly.AI llms.txt experiment (Feb 2026) · Google, “Optimizing your website for generative AI features” (May 2026) · Aggarwal et al., GEO (Princeton, 2023) · UK CAP Code section 3.7. Figures reported as published by each source; this page is re-checked when major new evidence lands. Corrections welcome at hello@modemarketing.ai.
The AI Visibility Check measures what AI systems currently see and say about your business, benchmarked against your three closest competitors, with the top three fixes. No guarantees · because you just read why nobody honest can offer one.
AI VISIBILITY CHECK · £95What GEO can and cannot do: the full, honest explanation.