AI ranked stocks are shares ordered by a model's scoring rules. The order can help you decide what to research first, but it has meaning only alongside the covered universe, forecast horizon, inputs and calculation method. A high rank alone does not establish a likely return, a low-risk investment or a suitable portfolio holding.
In my analysis of 383 equities from an archived iPulse AI scoring cohort, six names appeared in both the one-year and five-year top tens. Changing the sorting rule mattered even more: only two members of the one-year score-based top ten also appeared in the top ten sorted by annualized return scenarios. Those are comparisons of historical model outputs, not subsequent investment results.
The practical question is therefore broader than which company is first. What did the ranking reward, what did it leave out, and would the shortlist change if you asked a different question? Below is the named example, followed by a method you can use to inspect another ranking.
I am the founder of iPulse AI, and this framework is published on our own site. The worked example uses our archived records; it is not an independent comparison of providers or evidence that our rankings outperform another service. It supports research, not a recommendation to trade any named company.
An archived AI ranked stocks example at two horizons
I used all 766 original equity records from iPulse AI's Batch 7 quantitative scoring cohort: 383 companies, each with a one-year and a five-year record. The reference closes are dated 18 September 2026, the stored snapshot date is 20 September, and scoring ran on 21 September. I extracted the records on 5 October 2026. A fresh extraction does not make the underlying forecasts current.
For each horizon, I sorted the stored dividend-inclusive Consensus Score from highest to lowest. Stable asset IDs broke ties. The rank below is recalculated within these 383 equities, rather than copied from a stored rank that may cover a broader asset universe. There is no company, country, sector, positive-return or rating filter after selecting the cohort.
The records use the non_adjusted voice mode, scoring algorithm v2.3 and scoring architecture v7.4. These identifiers describe the quantitative scoring layer; this is not a test of the separate research synthesizer or a weighted portfolio strategy. Parentheses show the stored score, on the platform's bounded scale, rather than a return percentage.
| Rank within 383 equities | One-year score shortlist | Five-year score shortlist |
|---|---|---|
| 1 | Intuit (452) | Stellantis (373) |
| 2 | Adyen (416) | HDFC Bank (362) |
| 3 | Accenture (412) | Proximus (360) |
| 4 | Boston Scientific (410) | Intuit (350) |
| 5 | Ping An Insurance (409) | Ping An Insurance (349) |
| 6 | Adobe (408) | Boston Scientific (347) |
| 7 | RELX (397) | Adobe (335) |
| 8 | Tata Consultancy Services (395) | Tata Consultancy Services (330) |
| 9 | HDFC Bank (382) | NIKE (326) |
| 10 | Edenred (368) | Power Grid Corporation of India (326) |
Source: iPulse AI original Batch 7 equity scoring records, September 2026; extracted 5 October. Company names are shortened for readability. Scores are archived research outputs, not current prices, recommendations or realized returns.
Six names recur: Intuit, Boston Scientific, Ping An Insurance, Adobe, Tata Consultancy Services and HDFC Bank. Four places in each top ten belong to different companies. That difference does not show that one horizon is more accurate. It shows that a shortlist depends on the question being asked.
This is not just a top-ten boundary effect. The two score lists share two of their top five, 14 of their top 20 and 39 of their top 50. Those counts are descriptive checks within the same cohort. They do not estimate how stable rankings are across months, engines or market regimes.
There are score ties elsewhere in the full lists, so I preserve a deterministic tie rule. At the top-ten boundary, the tenth and eleventh one-year scores are 368 and 367; the corresponding five-year scores are 326 and 325. A one-point separation creates an ordered list, but it does not establish a meaningful economic advantage for the tenth company over the eleventh.
A high score and a high return scenario answer different questions
Next I kept the population, dates, horizons and dividend basis fixed, and changed only the sort field. Instead of Consensus Score, I ordered the records by their stored annualized mean total-return scenarios. I used the same asset-ID tie rule.
| Shortlist size | Shared names: one-year score vs return sort | Shared names: five-year score vs return sort |
|---|---|---|
| Top 5 | 0 | 0 |
| Top 10 | 2 | 1 |
| Top 20 | 7 | 10 |
| Top 50 | 25 | 24 |
These are membership overlaps, not accuracy statistics. A total-return scenario includes modeled dividends; it remains a forecast in the instrument's native currency, with no common-currency portfolio conversion. Sorting does not establish that any projected gain will occur.
The result gives a concrete reason to read the column header. In this cohort, selecting the ten largest one-year return scenarios produces eight names absent from the score-based top ten. At five years, nine are absent. Neither list becomes superior merely because its figures look larger.
The iPulse AI Consensus Score methodology explains why score and return can diverge. It combines forecast return with considerations such as dispersion, historical volatility, path instability and event-risk pressure. It is a bounded research signal, not a probability of success. The article's extraction checks the stored scoring configuration rather than assuming that a current documentation page defines every past edition.
Consider a hypothetical comparison: one company has a larger average forecast, but its supporting views disagree widely; another has a smaller average with less dispersion. A composite score may put the second company first. That is a consequence of what the formula rewards. It is not evidence that the second company's stock will earn more.
When a ranking changes, distinguish new business evidence from a changed sorting rule, updated reference price, revised model or different population. Those changes can all move a name without implying the same economic story.
Where to find ranked stocks, and what each list means
If you want an updated shortlist, use the provider's ranking page and inspect the displayed date and selected filters. The historical table above is an explanation you can revisit; it should not be substituted for today's list.
Danelfin's US stock ranking describes its AI Score as a ranking associated with the probability of outperforming the market over the next three months. Its methodology identifies the S&P 500 Total Return index as the US benchmark. That is a different target from a one-year absolute return scenario. The vendor's explanation is useful for understanding its target; this article does not independently validate its probabilities or reported strategy performance.
Seeking Alpha's Quant Ratings guide describes sector-relative comparisons covering value, growth, profitability, momentum and earnings revisions, with underlying metrics and rating history. That is quantitative factor research. Its rating should not be confused with a generative AI report, a specific price target or a directly comparable score from another service.
iPulse AI's Rankings provides a route into its model-generated research signals. Match the asset class and horizon, then open the evidence behind a candidate. Scores and labels are research outputs. Our seven-tool stock research comparison covers the broader workflow differences and other providers.
The first selection step is to write down the provider's actual target. A forecast of absolute gains, a prediction of benchmark outperformance and a sector-relative factor rating can all be useful, while answering different questions. Converting their scores to the same visual scale does not make their methods equivalent.
Five checks before relying on an AI stock ranking
1. Define the universe before interpreting the position
Record the denominator. A rank of 10 out of 383 is different from 10 out of every listed company worldwide. A stock can move up when other companies leave the covered universe, even if its own score stays unchanged.
Check geography, exchanges, instrument types and eligibility exclusions. A US-only ranking cannot directly answer a question about global equities. A dividend filter or an industry screen changes the population. Keep both the original universe and the filtered shortlist when you compare them.
2. Match the horizon and the return basis
A three-month market-relative model and a five-year absolute scenario require different evidence and patience. Write down the start date, end date, reference price and whether the return includes dividends. Separate annualized figures from cumulative horizon figures.
For the detailed arithmetic, see how to compare AI stock forecasts using CAGR and total return. Annualization helps compare units; it does not eliminate the different uncertainties of a short and a long forecast.
3. Inspect the underlying inputs and opposing case
A score is a compression of evidence and modeling choices. Open the explanation and identify the business assumptions it depends on. Which filing supports the margin claim? What needs to happen for the forecast endpoint to be plausible? What evidence would contradict it?
Also distinguish the number of research perspectives from the number of independent models or independent facts. Several outputs can share a model family, prompt pattern or source. Agreement may therefore repeat a shared blind spot. An opposing case is useful when it identifies a mechanism that the average conceals.
4. Check the date and keep the original edition
A ranking page's update date and a forecast's evidence date are separate fields. A new price can change remaining upside to an old endpoint without generating a new business thesis. A model update can also change the score without a new corporate announcement.
Preserve the original list, method version and assumptions before comparing a later edition. The Forecast Library explanation describes why retained forecasts matter. For a later assessment, compare what was available at the time with what happened afterward; do not replace the original record with a revised one and call it a successful prediction.
5. Separate a research shortlist from a portfolio
Ten ordered names do not define investment weights, position limits, diversification or a rebalancing schedule. Different companies can depend on the same economic driver. Equal weighting would be an additional strategy choice, not something established by the rank itself.
Investor.gov's stock guidance explains that stock prices can fall and investors can lose money. It also explains diversification across holdings. A model's confidence cannot remove business risk, market risk or your own need for liquidity.
A short ranking inspection worksheet
Before researching a name, save these seven items together: provider, snapshot date, covered universe, horizon, sorting field, score definition and the link to the underlying report. Then add a sentence stating the main thesis and another stating what would invalidate it.
Repeat the screen with one controlled change, such as horizon or sorting field. Count shared names rather than relying on an impression that the lists look similar. If the shortlist changes substantially, read the explanation for the differences. Do not select whichever version tells the most attractive story.
For a hypothetical rank comparison, suppose a company moves from twelfth to eighth while its score is unchanged. That movement alone cannot establish improvement in its business: four companies above it may simply have declined or left the universe. Check the score, inputs and membership before assigning a cause.
For later review, record the outcome measure before seeing the result. A benchmark-relative forecast needs a defined benchmark and matched period. A portfolio test needs weights, costs and a trading rule. A probability claim needs calibration evidence. This article supplies none of those performance tests; it explains how to preserve the information required to design them honestly.
What this example establishes, and what it cannot
The extraction contains no duplicate record IDs, no duplicate company-horizon pairs and no missing score or return fields used in these comparisons. Every company has both horizons. The membership checks cover top-five, top-ten, top-20 and top-50 cutoffs, with ties resolved by stable IDs.
The finding is narrow: horizon and sorting field can materially change a shortlist in one archived scoring cohort. It does not prove that this degree of change is typical, that rankings are useless, or that any named stock is attractive today. It also does not explain each company's business prospects. Those require separate, current research.
The strongest use of AI ranked stocks is as an organized entry into evidence. Start with the list, restore its definitions and dates, and inspect what would change the conclusion. To apply the worksheet, open iPulse AI Rankings, choose the population and horizon you intend to study, and follow one candidate into its underlying research.
Sources and evidence dates
First-party calculation: all 766 active original Batch 7 equity records in the selected non_adjusted voice mode, covering 383 companies at one and five years. Reference prices: 18 September 2026; snapshot: 20 September; scoring: 21 September; extraction and membership analysis: 5 October. Reproduction uses stored scores and returns, stable asset IDs, a fixed descending sort and the stated population filters.
Provider documentation and Investor.gov were reviewed on 5 October 2026. Their product descriptions explain methods, not independently verified performance. The cover is an original AI-generated editorial illustration of ranking inspection; its papers and screen are conceptual, not the source data for either table.
