AI Market Research: How moneyheap Analyzes a Stock

AI market research at moneyheap is a chain of small, checkable steps, not one model guessing. Market data and fresh news go in; a model scores how the evidence leans over three horizons and how strong that evidence is; then it writes down what it couldn’t know. This post follows one real stock, AppLovin (APP), through 5 November 2026 to show each step.

What does moneyheap’s AI market research actually do?

It answers three questions about a stock: what changed, why it matters, and how confident the analysis is. Each answer is built from public market data and news, scored on a fixed scale, and stored with its sources, so you can check the reasoning behind every number.

Three processes produce that research. Each one feeds a different part of the app:

Process	When it runs (New York time)	What it writes
Ticker analysis	weekdays at 18:00 for every followed stock; 9:45, 11:45 and 16:45 for trending stocks	sentiment and confidence for 3 horizons, a summary, the news behind it
Market Stories	every hour	stories built from scored news articles, with the tickers each one affects
Research agent	weekdays at 10:20	the Daily Standout and the shortlist behind it

Our methodology page is the formal reference: what the outputs are, which sources they use, and what the service doesn’t do. This post is the walkthrough. I’m writing it because “AI stock analysis” can mean anything from a chatbot with no data to a black box with a score, and I’d rather show exactly what ours does.

What data goes into an analysis?

Each analysis starts from three things: the stock’s prices and trading volume for the last 14 trading days, its previous analysis, and a fresh web search for news from the last 14 days. The model may use only those, so every claim in the result traces back to a price or a linked article.

But first, something has to put a stock in front of the model. For APP on 5 November, that was attention. AppLovin had reported third-quarter results the previous evening: revenue up 55% to $2.185 billion and fourth-quarter revenue guidance of $2.365 to $2.395 billion, according to its press release. By 9:45 the next morning, our trending check saw it at #2 in finance searches and #8 on social media, with 90 mentions against 3 a day earlier.

The trending check looks at five sources: social media mentions, finance search trends, the previous session’s biggest gainers and losers, the tickers in recent Market Stories, and the research agent’s biggest movers. A stock qualifies if it appears in two or more sources, makes a large move on real volume, has a sudden spike in mentions, or is one of the agent’s biggest movers. APP qualified twice over, on two sources and on the spike in mentions, and got a fresh analysis at 9:47.

How is the news checked and dated?

The model searches the web itself, and every news link it lists must be the exact URL its search returned. Each article keeps its source and publication date; when a page shows no date, the app labels it “Recent” instead of guessing. Pages the model read but didn’t list appear under “Also searched:”.

APP’s 9:47 analysis listed 10 articles from 4 and 5 November, covering the results, the earnings call, analyst reactions and the premarket move. They appear under Recent News in the app.

Market Stories works from the other direction. Every hour it reads business news feeds, keeps articles from the last 24 hours and scores each from 0 to 100 for likely market impact; opinion pieces and listicles score low. Articles scoring 75 or more are grouped into stories, one event per story, and the write-up may use only what those articles say.

The research agent adds one more rule: every name on its shortlist must be confirmed by at least two independent news sources plus market data.

How are the three horizons scored?

Each analysis scores three horizons: Short Term (7 days), Mid Term (1-3 months) and Long Term (6 months). Each gets a Sentiment Score from 0 (very bearish) to 100 (very bullish) that describes how the evidence leans. The label follows the score: 40 or below is Bearish, 41 to 59 Neutral, 60 or above Bullish.

The label is set by our code from the score, not by the model, so the two can never disagree. We learned that the hard way: early on, the model sometimes called the same 58 “Bullish” one day and “Neutral” the next.

The score bands mean the same thing for every model we use:

Score	What it means
50	no edge either way
41-59	a slight lean (Neutral)
60-69 (or 31-40)	the positives (or negatives) outweigh the other side
70-79 (or 21-30)	a fresh, concrete catalyst with numbers, and the price already reflects it
80-89 (or 11-20)	several strong independent facts point the same way and little stands against them
90+ (or 10 and below)	exceptional; almost never

APP’s 9:47 analysis came out at 66, 70 and 66 for the three horizons, all Bullish, with confidence of 62, 67 and 58 (mid on each). For what those labels do and don’t mean, see bullish vs bearish.

What does the confidence score mean, and when is it low?

Confidence measures how strong and consistent the evidence behind a score is. It is not the probability that the score turns out right. A 76 means several recent sources agree and the market has confirmed them; it does not mean a 76% chance the stock goes up.

The bands:

• Below 50 (low): thin, old or conflicting evidence, or the read hinges on an event still ahead.
• 50 to 64 (mid): the evidence leans one way, but important points cut against it.
• 65 to 74 (mid): several recent sources about this company agree, and nothing serious contradicts them.
• 75 to 89 (high): all of that, and price and volume already confirm it.
• 90 and above: almost never.

APP shows how this works in practice. At 9:47, a quarter of an hour into the session, the 7-day confidence was 62: the results were strong, but most of the jump had happened before the market opened, on little trading. By 10:27, the research agent saw the stock up 31.7%, with about 9.9 million shares traded in the first hour, nearly twice a normal full day. The news hadn’t changed much. What changed was that the market had now backed it with real trading, so the agent’s own 7-day read came out at 80 with 76 confidence (high). Its 6-month confidence stayed at 55, because the stock was still below its average price of the past 200 trading days and far under its high of the past year.

Why not call it a probability? Because we checked. When we compared earlier analyses with what prices did afterwards, the more confident calls were not right more often. Confidence tells you how solid the evidence is today. It can’t tell you what the market does next, and the risk disclosures say the same.

What can’t the analysis know?

A lot, and we write it down. The research agent files a list of limitations with every Daily Standout, and each stock on its shortlist comes with a note on what would change the view. Prices are snapshots, news can be wrong or syndicated, and a scheduled event can move everything after the scores are set.

On 5 November the agent’s list had nine items, including:

• Prices were a snapshot from about 10:27 on a very volatile day, so gains and losses could grow or shrink by the close.
• Adjusted profit figures leave out real costs such as stock-based pay, so reported profit can look different.
• Friday’s jobs report and the next week’s inflation data could move the whole market after the scores were set.

For APP specifically, the case against was spelled out too: the stock was still about 52% below its high of the past year, one large bank kept a neutral rating, and management gave no firm date for its new ad models. The agent noted that the view would weaken if the stock gave back most of the day’s jump or analysts began cutting estimates. In the app, this appears under What would change the view.

None of it uses anything about you. The analysis of APP is the same for everyone who follows it, and it never looks at your holdings, goals or circumstances.

How does the Daily Standout narrow the market to one stock?

The research agent works as a funnel: a broad universe, a size and direction filter, a fresh-catalyst filter, quality checks, then ranking. It records the count at each step and explains every cut, so the Daily Standout comes with a written account of how it was chosen.

On 5 November 2026, the S&P 500 was down about 2.8% and the Nasdaq about 4.5% by mid-morning, as chip and AI data-center stocks sold off. The funnel that day:

Step	Tickers left	What it kept
Universe	405	most-discussed on social media, biggest gainers and decliners, most active, companies in the day’s news
Size and direction	32	US companies worth $10 billion or more, rising 3% or more
Fresh catalyst	14	a company-specific reason from the last day, confirmed by 2 or more sources
Quality checks	9	after checking the size of the beat, one-time items, trend, momentum, volume and peers
Shortlist	7	one standout and six alternatives

APP ranked first: the largest gain among large US companies on a sharply down day, results and guidance above estimates, and the heaviest trading relative to normal. The universe size changes daily; on 2 November it was about 660. You can read the whole day, including the shortlist and the trending list, on the stock market today page for 5 November 2026. How the agent itself is built is in how our Daily Standout research agent works.

Momentum describes what already happened. A stock that stood out one morning can reverse the next day.

Frequently asked questions

Which AI models does moneyheap use?

Ticker analyses run on Google’s Gemini by default and on Anthropic’s Claude Sonnet 5.5 for the trending refresh. Both get the same instructions and the same score scale, and each analysis stores which model wrote it. The research agent behind the Daily Standout runs on Claude.

How often is an analysis updated?

Every followed stock gets a new analysis on weekday evenings. Stocks that start trending get a fresh one during the day, unless they were analyzed in the last 12 hours. Market Stories checks the news every hour, and the research agent runs once each weekday morning.

Can I check the sources myself?

Yes. Every analysis lists its news with links, sources and dates, and the public APP page shows the summary and the news behind it. On the Daily Standout, the funnel and the limitations are shown with the shortlist. You can also ask Claude about any stock’s analysis over MCP.

How we wrote this

This post describes moneyheap’s analysis as of 6 November 2026, based on our code (the analysis prompts and score scale, the news and trending pipelines, the validation rules) and the internal documentation for Market Stories and the research agent. The APP example comes from our own data for 5 November 2026: the ticker analysis generated at 9:47 New York time, that day’s trending list, and the research agent’s shortlist with market data as of 10:27. AppLovin’s figures were checked against the company’s press release of 4 November 2026. The comparison of confidence with later price moves is our internal check of earlier analyses, not a measure of future results. Scores are model opinions, not guarantees, and prices are intraday snapshots. We left out vendor names and internal identifiers.

moneyheap is a market research tool. This post is for information and education only. It is not financial advice or a recommendation to buy or sell any security. Do your own research.