AI visibility is not a stable state: according to AirOps (2026), only 30% of brands present in one AI answer are still present in the next, for the same query. Without structured proprietary signals, AI invents your identity from fragments — often your competitors'. AI Governance is the maintenance contract that protects the asset you've built.
It is a structured program of recurring audits that protects a brand's visibility in generative AI answers. It is designed for brands that are already visible — those that have built a measurable AI presence and want to preserve it against stochastic erosion. Unlike growth offers, it doesn't aim to raise the KVS score, but to prevent its fall. Six pillars are audited at mandated frequencies.
When a brand emits no structured signals, AI doesn't stay silent: it fills the gaps with whatever data is available — competitors', trade media's, forums'. That is the third-party narrative: your brand as told by others, in the only space where your future clients now look.
of brands present in one AI answer are still present in the next answer to the same query. Other measurements find more stability: it depends on the query and the engine.
AirOps & Kevin Indig, 2026 · methodology
of pages cited by ChatGPT were updated within the past 12 months. Refreshing matters more than novelty: measured by original publish date, the share drops to around 40%.
AirOps; Seer Interactive, July 2026 · methodology
average concentration of citations by source domain (Gini coefficient). A handful of domains captures most citations, yet roughly two thirds of cited sources change from one day to the next.
University of St. Gallen, 2026 · methodology
Each pillar answers a documented erosion risk. Frequencies draw on measurements published in 2026; sources and limits are detailed at the bottom of the page.
Figures, statistics, prices, dates. According to AirOps (2026), a page not updated for more than three months is over three times as likely to lose its AI citations. Quarterly updates keep your critical data inside that window.
Name, address, phone, descriptions, expert biographies. Any cross-source divergence reads as a low-reliability signal to RAG models. Every divergence found is corrected at the source, not only on your own site.
Organization, Person, VideoObject, ImageObject, sameAs property. The technical backbone evolves with the models; the yearly audit catches regressions and new types to integrate.
The heart of governance. KVS score recalculated every month: 7 brand runs, 8 source runs, standard error under 0.10. Jaccard and RBO indices to catch volatility before it becomes erosion. Run counts derived from Schulte et al., University of St. Gallen (2026).
Reddit, specialist forums, YouTube. Their weight in citations varies sharply by engine: very high on Perplexity, marginal on Gemini (Tinuiti, 2026). That's where the third-party narrative forms — so that's where it must be watched and corrected.
Adopted by Anthropic, Cloudflare, Yoast and AIOSEO in 2026, not yet validated by Google or OpenAI. Offered only to content-heavy sites (authority blog, documentation, complex e-commerce), with its "unproven" status clearly flagged.
The annual audit diagnoses. The recurring plan maintains. Both protect against the third-party narrative — private banks, clinics, professional firms and regulated professions first in line, compliance validation workflow included. Pricing on quote.
A complete snapshot at a point in time + a 90-day action plan. For brands that want a diagnosis with no commitment.
The asset-maintenance contract. All 6 pillars audited at their mandated frequency, monthly monitoring, drift alerts.
AI Search Advisory and Managed GEO Growth have a growth objective: they raise your KVS score. AI Governance has a conservation objective: it protects the acquired asset against stochastic erosion. The two can coexist for a mature client, or follow each other once the growth phase is done.
The one-off audit is enough for a snapshot ahead of a structuring event (fundraising, M&A, redesign, crisis) or to validate an internal plan. The recurring plan is necessary to catch and correct drift as it happens: AI visibility moves every day, a yearly audit leaves an 11-month blind spot.
The standard is widely adopted in 2026 (Anthropic, Cloudflare, Yoast, AIOSEO) but has not been validated as a citation factor by Google or OpenAI. Implementation cost is near zero, the gain plausible but unproven. We offer it only to content-heavy sites, with the "unproven" status clearly flagged. Distinguishing proven / anticipation / discarded is itself a signal of expertise.
An automatic alert triggers as soon as a drift of more than 10 points is detected: diagnosis within 5 business days, corrective plan within 10. The goal isn't zero fluctuation (AI visibility is stochastic) but zero undetected structural erosion.
Every study cited is correlational: it describes what engines cite, not what causes a citation. We use these studies to calibrate audit frequencies, never as guarantees of results.
Metric: share of brands present in one AI answer that are still present in the next answer to the same query (30%); about 20% stay visible across five runs. The report also links going more than three months without an update to an over threefold higher risk of losing citations.
Sample: 45,000+ citations across 800 queries. Scope: AirOps' own query set, mostly English-language.
Limit: other measurements find far stronger persistence. Across 22,295 answers (ChatGPT, Perplexity, Google AI Mode), Analyze AI found that a brand present one week is still present the next about 83% of the time.
Metric: share of cited pages whose last update is under 12 months old (over 70% at AirOps, pages cited by ChatGPT). This is not the age of the page.
Cross-check: Seer measures 75% of cited pages updated within the year across ChatGPT, Gemini and Perplexity, but only about 40% first published within the year among the most-cited pages.
Limit: correlation. Maintained pages are also the pages someone cares about; Seer could only date part of its sample.
Metrics: Gini coefficient of citation distribution by source domain (mean 0.715; from 0.671 on Perplexity to 0.782 on Google AI Mode); Jaccard similarity and RBO (p = 0.9) between answers; standard error by number of runs (under 0.10 at 7 runs for brands, 8 for sources).
Sample: 4 verticals (telecoms, real estate, sporting goods, consumer electronics), 8 prompts each, 4 engines (ChatGPT, Gemini, Google AI Mode, Perplexity), daily collection from 24 January to 20 March 2026, plus a 10-run simultaneous dataset.
Scope and limits: Swiss-German market, German-language prompts, Swiss servers. The Gini measures cited domains, not brands; it is a snapshot of concentration, not a measure of mobility. Code and data published by the authors.
Metric: share of AI citations coming from social media and community platforms, by engine. Example: 31% of Perplexity's January 2026 citations, 24% from Reddit alone; very low share on Gemini, Meta AI and Copilot.
Limit: published estimates vary widely by method and engine; no single figure holds for "AI" in general.
Sources last verified: 7 October 2026.
Start with a 20-minute call to identify the right plan for your situation. No commitment · decision within 48 h · 100% confidential.
Book a call