A 63-year-old widow who still works part-time asked an AI chatbot whether she should start a survivor benefit on her late husband's record now, let her own retirement benefit keep growing, or reverse the order if the numbers favoured that. The chatbot answered that deemed filing prevented her from claiming one benefit while delaying the other. The explanation was fluent, detailed and confident - it named a real Social Security rule and applied it to her case without hedging - and it left her with a claiming strategy built on a rule that does not govern the benefit she was asking about.
1 Answer
Expert: Gerelyn Terzo, Personal-finance journalist, 24/7 Wall St. Deemed filing is a real rule. The chatbot attached it to the wrong benefit, and the difference is worth thousands of dollars a year for life, plus a permanent reshaping of tax brackets and survivor benefits. Deemed filing generally forces someone who applies for retirement or spousal benefits to apply for both at once - it exists so a claimant cannot take one and delay the other to game the system. It does not apply to survivor benefits. Social Security explicitly permits an eligible widow or widower to begin survivor benefits independently of personal retirement benefits, which is precisely the door the widow was asking about. That independence creates two legitimate paths. She can claim a reduced survivor benefit first and switch to a higher personal retirement benefit as late as age 70; or she can claim a reduced personal retirement benefit first and switch to a higher survivor benefit once she reaches survivor full retirement age. The right answer depends on both benefit estimates, her age, her late husband's claiming history and her current earnings - which is exactly the kind of case-specific arithmetic a chatbot was asked to perform and could not. Her part-time job is not a footnote. Survivor benefits are subject to the retirement earnings test before full retirement age. In 2026 someone under full retirement age for the entire year can earn $24,480 before withholding begins, and Social Security withholds $1 in benefits for every $2 earned above that limit. The strategy can still work, but the paycheck has to be inside the comparison. Why a confident wrong answer is expensive here: a claiming decision is close to irreversible. Social Security generally allows an application to be withdrawn only within 12 months of approval, the benefits received must be repaid - including certain payments made to family members - and the withdrawal is generally allowed once. After full retirement age a recipient can only suspend retirement benefits to earn delayed retirement credits until 70, and that is not a full reset. On top of that, the 2026 cost-of-living adjustment of 2.8% is applied to whatever benefit the claiming decision produced: start from a smaller base and every future increase starts there too. The same conversation pattern shows up elsewhere in Social Security, which is full of rule pairs that sound nearly identical and produce opposite answers. Remarrying at 60 or later generally does not eliminate eligibility for survivor benefits on a deceased spouse's record, while remarriage generally does end divorced-spouse benefits on a living former spouse's record. The Windfall Elimination Provision and Government Pension Offset - which once reduced benefits for many teachers, police officers, firefighters and other public workers - were repealed by the Social Security Fairness Act for benefits payable beginning in January 2024, so a chatbot working from older material can still recite reductions that no longer exist. And for the retirement earnings test, wages and net self-employment income count before full retirement age, while pensions, IRA withdrawals, dividends, interest and capital gains generally do not; treating every dollar entering a retiree's household as 'earnings' produces a wrong claiming recommendation. The useful division of labour is narrow and clear. Let a chatbot translate Social Security jargon, surface rules worth checking and help build a better list of questions. Then take that list to the source: pull retirement estimates and earnings history from your my Social Security account, ask the agency for survivor-benefit estimates at different claiming ages, and compare both paths with part-time earnings included - or work it through with a fee-only adviser. AI can help draw the map. It should not choose the road. The FINRA Investor Education Foundation's National Financial Capability Study found 20% of U.S. adults say they would be interested in receiving financial advice from AI, which makes this a mainstream failure mode rather than an edge case: a wrong answer delivered hesitantly invites another question, but a wrong answer delivered with confidence becomes a plan. Source: https://247wallst.com/personal-finance/social-security/2026/08/08/she-asked-an-ai-chatbot-when-to-claim-social-security-it-sounded-certain-and-it-was-wrong/
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More from this topic
Across 121 money questions covering debt, mortgages, pensions and tax - each run five times, more than 10,000 responses in total - the models gave answers that were wrong or incomplete 57% of the time, and presented them as settled guidance. On the hardest multi-step questions the failure rate reached 88%. Gemini 3.5 Flash and Claude Haiku 4.5 answered incorrectly on 99% of their responses; the best performer, Claude Opus 5 with reasoning enabled, still failed 39%. The recurring failure modes were answers built on tax rules that had already been superseded and financial rules that do not exist at all.
Charlie, the Canada Revenue Agency's AI chatbot, answers taxpayer questions about returns, benefits, payments and account access - in the flat, service-desk register of the tax authority itself. Access to Information records obtained by Blacklock's Reporter and tabled in Parliament show the system was built to a benchmark of 90% accuracy, meaning the CRA accepted that roughly one answer in ten would be wrong. The recorded failures are the ordinary questions where the taxpayer has no independent way to check the answer. Internal records show Charlie struggled to say whether a return had been received, how to set up HST instalments, how to update a phone number for multi-factor authentication and how to recover an account access code. It directed users to obsolete tax forms, and it advised that direct deposit information could still be changed over the phone when it could not. Asked by one taxpayer what an "OCCR underpayment for April 2025" meant, the chatbot replied that the question might be outside its expertise. A review dated Oct. 29 found that only 36.4% of users who provided feedback were satisfied, while 63.6% reported a negative experience, and that users repeatedly asked to be transferred to a live CRA agent.
A general-purpose AI chatbot recommended a Monaco-friendly tax strategy to a UK employee based in Croydon - advice that was useless for him, because the model ignored UK tapering allowances and the contributions he had already made. It is the worked example the Financial Times reported alongside the FCA's Mills Review (published 6 July 2026), which examines consumers 'routinely turning to general-purpose AI tools for everyday budgeting, saving and investment tips'. The review asks whether AI systems could deliver services 'functionally equivalent to regulated activities while remaining outside the regulatory perimeter' - including agentic AI that compares products, rebalances portfolios or executes trades. Consumer trust is running ahead of performance: a Lloyds study found 28 million UK adults used AI for personal-finance questions in 2025, and Fidelity data cited by the FT showed 36 percent of 18-to-34s turning to it for investment ideas.