US Equities × AI-Assisted Research

I asked Fable 5 to pick stocks for a wobbly market. Then I checked the logic.

10 min read · June 2026

The S&P 500 had just snapped a nine-week winning streak. May's jobs number — 172,000 versus an 85,000 consensus in Reuters' report — pushed the 10-year yield back toward the mid-4% range and lifted market-implied odds of a December Fed hike to roughly the high-60s to mid-70s, depending on the exact timestamp and source. Semiconductors led the Friday, June 5 selloff.

In that backdrop, I ran an experiment: I gave Fable 5 — Anthropic's top-tier AI model — a detailed buy-side analyst prompt and asked it to find five US stocks likely to beat the S&P 500 over the next 12 months. Then I sanity-checked the logic, timestamps, and scenario math. No vague requests, no "just give me some ideas." I imposed a structured investment framework, live data requirements, explicit probability weights, and a built-in bias-check.

The goal wasn't to outsource my judgment. It was to stress-test my thinking with a tool that can't talk itself into a stock it wants to own — and then check where the model's own numbers needed adjustment.

Here's the full methodology, the prompt I used, Fable 5's output, and the corrections needed to make the scenario table internally consistent.

The macro backdrop that shaped the picks

Before choosing stocks, Fable 5 grounded the analysis in the current macro regime. The key variables as of June 10, 2026:

Variable	Current Reading	Direction
10-Year Treasury Yield	4.55%	Rising — strong jobs data
Fed rate-hike odds (Dec 2026)	~65–75%	Up sharply after strong jobs data
VIX	~19	Off highs but elevated vs late-2025
S&P 500	~7,4xx	Post-selloff bounce, still cautious
Regime label	Sticky inflation + rate-hike risk + geopolitics	Growth-unfriendly

That regime — sticky inflation, the prospect of hikes under new Fed chair Kevin Warsh, and an energy supply shock from Strait of Hormuz disruptions — changes what beats the index. Narrow AI momentum is already priced. The question is what isn't.

The prompt: what I actually asked

Most AI stock prompts are terrible. They ask for "good stocks" and get back a list of the five most-talked-about companies in the training data. I wanted something better. I structured the prompt around how a buy-side analyst actually evaluates a stock.

Prompt Framework (condensed)

Role: You are a buy-side portfolio manager and analyst with 10+ years of experience. I want US stocks likely to generate excess returns over the S&P 500 in the next 12 months.

Hard rules:

Use web search to pull today's live data — price, earnings, guidance, consensus, macro. Cite every source with a date. Mark anything unavailable as "unknown." No guessing.
Define what "outperform" means — we use S&P 500 excess return as the target.

For each stock, analyze:

Fundamentals: revenue/EPS growth, margin trend, FCF, balance sheet
Valuation: current P/E, EV/EBITDA vs historical range, peers, and growth rate — with numbers
Catalysts: specific events in the next 12 months and their expected timing
Macro sensitivity: beta to rates, USD, economic cycle
Positioning/sentiment: is the thesis already priced in?

Probability and expected value (required):

Bull / base / bear scenarios, each with a 12-month price target and probability (sum to 100%)
Probability-weighted expected return and maximum drawdown in the bear case
One or two explicit invalidation conditions — what would prove this thesis wrong?
Conviction: High / Medium / Low, with explanation

Bias checks: Reject any pick based purely on recent momentum. Exclude narrative-led stocks unless the fundamentals independently support the thesis. End each write-up with the most likely reason the recommendation is wrong.

That last rule — "end with why you're wrong" — is the one most investors skip. It is also the one that protects you from the most expensive mistakes.

Important caveat: I treated Fable 5's answer as a research draft, not a recommendation. The targets and probabilities below remain subjective inputs. Where market prices had moved or the article's numbers did not reconcile, I recalculated the scenario returns mechanically from the stated targets.

The five picks, ranked by conviction

Here is Fable 5's output, compressed into the key thesis and numbers. Full scenario analysis follows in the charts below.

UNH

UnitedHealth Group · ~$413

A large-cap healthcare turnaround candidate. Q1 medical care ratio came in at 83.9%, FY adjusted EPS guidance was raised above $18.25, and management signaled continued improvement. The stock still trades below its own historical multiple, but the expected-value math is now tighter after the rebound.

EV ~+2% Conviction: Medium

GOOGL

Alphabet · ~$364

Cloud growth and margin expansion remain the core fundamental support. The roughly $80–85B equity raise for AI capex created a temporary supply overhang, but the raise itself also validates the scale of AI infrastructure demand. Berkshire's reported $10B participation is a useful quality signal, though dilution risk needs to be watched.

EV ~+10% Conviction: Medium

BRK.B

Berkshire Hathaway · ~$488

$397B in cash earns more as rates rise. Q1 operating income +18% YoY. In a rate-hike or drawdown scenario, this is the portfolio's shock absorber — low beta, low rate sensitivity, buyer-of-last-resort positioning.

EV ~+6% Conviction: Medium

RTX

RTX Corporation · ~$182

Record $271B backlog, including $109B defense. Q1 adjusted EPS rose 21% year over year and management raised the 2026 outlook. Pratt & Whitney commercial aftermarket was up 19%. Two revenue engines — defense spending and air travel recovery — move independently of AI sentiment.

EV ~+14% Conviction: Medium

XOM

Exxon Mobil · ~$149 · Portfolio hedge, not pure alpha

This is explicitly a geopolitical hedge: if the Strait of Hormuz disruption persists and WTI stays elevated, XOM acts as insurance against the macro scenario that hurts everything else in the portfolio. Inclusion is conditional on the energy risk remaining live; if the geopolitical premium fades, this becomes less of an alpha idea and more of a commodity-cycle exposure.

EV ~+5% Conviction: Low-Med

12-month scenario returns — bull, base, and bear

Percentage price returns from current price in each scenario, before dividends. The index baseline assumption is roughly +3–5% under the rate-hike-risk framework, but this is an assumption rather than a forecast.

Scenario prices and probabilities are Fable 5's estimates; return percentages were recalculated using approximate market prices as of June 10, 2026. Not a guarantee of returns.

Probability-weighted expected return vs. bear-case drawdown

Risk/reward summary. Bars show probability-weighted expected price return before dividends. The line shows the absolute bear-case drawdown from the same current-price base. This is a framework, not a forecast.

Bear-case drawdown calculated from current price. Expected return is probability-weighted across all three scenarios and excludes dividends.

The scenario table in full

Ticker	Current price	Bull target / prob	Base target / prob	Bear target / prob	EV (price-only)	Bear drawdown
UNH	$413	$490 / 30%	$440 / 45%	$310 / 25%	~+2%	-25%
GOOGL	$364	$470 / 30%	$420 / 45%	$280 / 25%	~+10%	-23%
BRK.B	$488	$560 / 25%	$530 / 55%	$430 / 20%	~+6%	-12%
RTX	$182	$235 / 25%	$215 / 50%	$160 / 25%	~+14%	-12%
XOM	$149	$185 / 25%	$160 / 45%	$125 / 30%	~+5%	-16%

What Fable 5 explicitly excluded — and why that matters

The most instructive part of the exercise was not what made the list. It was what got cut.

Notable exclusion: MU (Micron Technology)

Micron's fundamentals are genuinely strong. HBM supply is fully contracted through end-2026, and fiscal Q1 revenue was $13.64B, up 57% year over year. HBM4 positioning for Nvidia's Vera Rubin platform keeps the long-term thesis interesting.

But the stock had already made an enormous move, had just plunged 13.25% in a single session on June 5, and Fable 5's captured analyst-target set showed an unusually wide range. That kind of spread reflects genuine valuation uncertainty, not analytical precision.

Fable 5's reasoning for exclusion: "I could not identify an edge beyond recent momentum. The probability-weighted expected value at the current price was approximately zero to slightly negative. The June 24 earnings print is a coin-flip in either direction. This is exactly the recency bias the prompt required me to reject." The pick was declined even though the underlying company is compelling.

The discipline signal: a good process should exclude stocks you want to own when the numbers don't support the entry price. That's harder than it sounds — and it's the step most retail investors skip.

The invalidation conditions

Each pick comes with explicit conditions that would break the thesis. Fable 5 was required to state these upfront — not as a footnote, but as a decision rule. A few of the clearest ones:

UNH: MCR worsens for two consecutive quarters, or the DOJ investigation produces a formal criminal charge. Either event breaks the turnaround thesis entirely.
GOOGL: Cloud growth falls below 40% for two quarters, or the equity raise forces additional secondary offerings that signal the capex program has grown beyond manageable FCF.
BRK.B: Abel fails to deploy the cash pile in a downturn — negating the optionality argument — or insurance underwriting deteriorates significantly from a major catastrophe year.
XOM: A durable ceasefire holds, Hormuz traffic normalizes, or oil risk premium unwinds. At that point, the hedging rationale becomes much weaker.

The bear case against the whole analysis

Fable 5 was also required to state the most likely reason the entire list is wrong. Here's the verbatim logic:

"The most likely failure mode is that the stagflation concern turns out to be overblown. If this week's CPI undershoots, Warsh opens dovishly at his first FOMC, and a Middle East ceasefire holds — markets will immediately price 'rate-cut revival + risk-on re-acceleration.' In that world, this portfolio's defensive tilt underperforms badly. The quality of the thesis depends on the macro premise staying intact. I assign roughly a 60% probability that it does — meaning there's a meaningful 40% world where the index simply beats everything on this list."

That's not a standard AI output. Most tools won't tell you there's a 40% chance the whole exercise was directionally wrong. Forcing that disclosure was the most valuable part of the prompt design.

What this experiment actually tells us

Fable 5 is not a stock-picking oracle. Even with web access, it has no proprietary data and no ability to feel the market. But with a rigorous enough prompt, it can do something useful: it forces structure onto a naturally emotional process.

The prompt required live data, explicit probabilities, stated invalidation conditions, and a self-written bear case. Most investors skip at least two of those four. The AI's value was not in the insight itself but in the discipline of following the framework without exception.

Practical takeaway: use AI for the framework, not the answer. The five stocks above are starting points for your own research — not a ready-made portfolio. Run the invalidation conditions against your own macro view. If your read on the Fed path differs from the baseline, the expected values shift significantly.

Data notes and limitations

Market prices and scenario return percentages were refreshed on June 10, 2026. Price targets and probabilities remain Fable 5-generated scenario inputs, not consensus estimates.
The Fed rate-hike probability is shown as a range because futures-implied odds moved quickly after the May jobs report and differed by timestamp/source.
Scenario probabilities are subjective estimates, not actuarial data. They should be treated as directional, not precise.
The S&P 500 "baseline" return of +3–5% is an assumption. If the index performs better or worse, the relative performance calculus shifts accordingly.
Even with web access, AI-assisted research is timestamp-sensitive. The June CPI print, Fed communication, and Middle East headlines could change the entire framework quickly.
Source checks used in this revision include Reuters on Fed-hike odds and Alphabet's equity raise, UnitedHealth's Q1 report, and RTX's Q1 release/backlog disclosure.

For educational purposes only. Not investment advice or a recommendation to buy or sell any security. All scenario returns and probability estimates are illustrative and subject to substantial uncertainty. Past performance does not indicate future results. Always do your own research and consult a licensed financial advisor before making investment decisions.