Back to News
AI reputation management after a hack: why engines skip your incident report

AI engines quote the page that holds the claim they are writing, and after a hack, that is rarely your own post-mortem.Ledger published its response to the CryptoBilis implant attack on 15 October.By the next morning, ChatGPT was quoting it, while Google still did not show it.That morning, we asked four AI engines 10 questions about the incident, and Ledger's post appeared as a source in 5 of the 40 answers. The other 35 did not cite it.An engine treats your incident report as one source among many and cites it only when it holds the exact claim the answer needs.AI reputation management after a hack starts with writing those claims into the page.A hack ends in a week. AI keeps telling the story for months.This column is part of Crisis Memory, ICODA's November series on what ChatGPT, Perplexity, Claude and Gemini say about crypto brands after a crisis, with insights from Vlad Pivnev, Anastasia Marmuzevich, Anastasia Shalepina and Artyom Abbasov.Did AI quote Ledger's post-mortem?Only partly: ChatGPT cited the 15 October post in 4 of its 10 answers, Perplexity in 1 of 10, and Claude and Gemini never did.ChatGPT sent 24 of its 34 citations to Ledger's own sites, including the 2020 breach notice and the 2023 Connect Kit report.Claude sent 1 of its 52 there and built its answers from BleepingComputer, Help Net Security, The Hacker News and Bitquery.Perplexity cited 232 sources across its 10 answers, and Ledger's post appeared in one of them.Gemini skipped web search in 6 of its 10 answers, and none of those answers mentioned the incident.How do AI engines pick a source after a hack?An engine cites the page that contains the claim it is writing, so every claim your post-mortem leaves out gets sourced from someone else.Ledger's answers show that rule in three claims.Claim in the answerSource the engine usedWhat we sawChip, OS and servers were not compromisedLedger's response postChatGPT: "Ledger says confirmed cases were devices redistributed through resellers in Southeast Asia, rather than a compromise of Ledger's servers, OS, or secure chip."How much was stolenSpecter, Bitquery, crypto media12 of 40 answers gave a figure, 9 named a source, none called it Ledger-confirmed. The post states no figure.When it happenedThe date on the page the engine citesTwo ChatGPT answers date Ledger's disclosure to 15 October: "On October 15, 2026, Ledger disclosed a supply-chain attack". Ledger's first statement came on 9 October.In the second row, Claude gave Bitquery's $102.0M across 324 wallets and Specter's $86M+, then added: "Ledger itself said it has not confirmed the total number of affected users or the total value lost."In the third row, the post is dated 15 October and mentions no earlier statement, although Ledger had spoken on 9 October and confirmed the implant on 10 October.Does a Google ranking mean AI search visibility?No: on 16 October, ChatGPT quoted Ledger's post while Google did not show it, so rank and AI search visibility need separate tracking.For the query "ledger hack", an old phishing-campaigns page still held #2, and the 15 October response was missing. Engines run their own retrieval, so a page can be quoted before it ranks.Google's AI Overviews diverged by market too: after the news broke, they vanished from brand queries such as "ledger" in the US, UK, France and Germany, and stayed in Malaysia and Indonesia, where they cited crisis coverage within one to three days."First hours, nobody recalls statement quality; they remember search results," said Anastasia Marmuzevich, Head of PR at ICODA.Old pages compete with the new one. ChatGPT cited Ledger's 2020 breach notice and 2023 incident report next to the new post.In the Crisis Memory prompts about Ledger, the 2020 leak appeared in 25 of 48 answers and the 2026 implant in 15.What should an incident page contain so AI quotes it?Write the page as a set of claims an engine can lift: dated milestones, confirmed numbers, who is affected and what users got back.Date every milestone: the first statement, the confirmation, and each update. A page that shows one date hands the engine that date.State what is confirmed and what is not, with numbers where you have them. Without them, engines fill the gap with Specter's $86M and Bitquery's $102M.Put the recovery in the same paragraph as the hack. In our Crisis Memory data, recovery came up in 33 of 34 Bitget answers that mention the hack, 30 of 38 for Bybit, and 11 of 45 for Kelp DAO.Turn the questions people ask into headings, such as "Were users compensated?" and "Is the device safe to use?", and answer each in the first sentence.Get one independent outlet to repeat your facts: 51 of Claude's 52 citations went to third-party sites.Check all four engines in the first week, because answers barely move afterwards. When we repeated the audit 24 days later, only 2 of 40 carried news from after 16 October."Treat the incident page as an answer sheet: every question an engine might get about the hack should have a dated, quotable answer on it," Anastasia Shalepina, AI SEO Expert, ICODA.Method: API versions of ChatGPT, Perplexity, Claude and Gemini with web search, US location, one run per prompt, 16 October 2026. Crisis Memory covers 296 answers across five crypto brands.🔍 Check if AI cites your brand → https://icoda.io/ai-visibility/
More crypto intelligence
- Kraken Investigating Issue With The Ink Funding Gateway
- Bitcoin enters bearish phase as demand shrinks, traders face 20% losses
- XRP Price Today: $0.95 Support Holds, Gaussian Channel Flags $0.84 as a Potential Cycle-Bottom Zone
- CryptoQuant says bitcoin is back in a bearish phase but has reached a ‘value zone’