Credible Roots / Case studies / heatpumpmath.com
Case study · SEO
The SEO differentiator was the methodology page
A consumer calculator in a crowded category, built on a live government data pipeline instead of national averages, with every assumption printed and every error logged in public. The first named client on this site, published with permission.
More than five years in personal branding. Runs atharvsankpal.com and works with founders and businesses in the US on their digital presence. Previously produced AI video for brands.
Published 2026-09-04
Written from our own engagement records and from the client site, which is public and can be checked against every claim here. How we write · About the author
The short answer
heatpumpmath.com answers whether a heat pump costs less to run than the heating system already in the house. Every comparable tool we found ran on national average energy prices, which is the main reason they disagree with each other. We built the site on a script that pulls residential electricity and gas prices from the U.S. Energy Information Administration's Open Data API v2 for 51 states including the District of Columbia, fixed to a December-to-February heating-season window because summer gas volumes collapse and fixed customer charges then inflate the per-unit price enough to bias every comparison toward heat pumps. Four assumptions are printed with their values, no winner is declared inside a 10% margin, six states are excluded from gas comparisons where the data will not support one, and the computed figures are released under CC BY 4.0. On 31 August 2026 we published a correction: the pipeline had been using the most recent monthly figures rather than the heating season, and fixing it cut the number of states where a heat pump wins on running cost under cold conditions from nineteen to four. That correction is on the site with both numbers side by side.
- The moat is the pipeline, not the copy. A competitor can rewrite the page in a day and cannot rebuild the data.
- The window was a decision, and the reason for it is published.
- The correction made the headline finding weaker, and it was published anyway.
- Everything here is checkable on the live site, which is the point of the case study.
On this page
Who they are
heatpumpmath.com is run by Ash Banerjee, a mechanical engineer by training, and answers one question for American homeowners: does a heat pump cost less to run than the gas furnace, propane, oil or electric resistance system already in the house. Credible Roots does the SEO. The engineering judgement, the constants and the editorial calls are the client's; the site structure, the search work and the publishing discipline are ours. They have given permission for this write-up.
The problem with the category
Search “heat pump running cost” and you get a page of calculators that disagree with each other. Almost all of them run on national average energy prices, which is a problem in a country where residential electricity costs roughly three times more in one state than another. A national average answer is wrong nearly everywhere, and confidently so.
That is the competitive situation, and it is a good one. When every result is the same shape and the same quality, the way to win is not to publish a better-formatted version of it. It is to be the only one holding a number the others do not have.
What we actually did
Read the SERP before designing the page
The results for the money queries were calculators, not articles. That settled the page type before any writing: the primary page had to be a working tool with the explanation underneath it, not an essay that mentions a calculator. Publishing the right content in the wrong format is one of the most common ways good work fails to rank, and it is decided at the wireframe.
Built the pricing on a live API rather than a spreadsheet
Electricity comes from the EIA Open Data API v2 (electricity/retail-sales, residential sector, monthly, cents per kWh) and gas from natural-gas/pri/sum (process PRS, monthly, dollars per thousand cubic feet), pulled by a script that lives in the project. That covers 51 states including the District of Columbia. It is more work than typing in averages once, and it is the entire reason the site can say something the competing pages cannot.
Fixed the heating-season window, and published the reason
Prices average December, January and February from the most recent complete heating season. The reason is stated on the methodology page: residential gas volume collapses in summer, fixed customer charges then dominate the per-unit price, and using the latest available month would systematically bias comparisons toward heat pumps. A defensible choice with the reasoning attached is worth more than a defensible choice on its own, because the reasoning is what a reader, a journalist or an answer engine can quote.
Printed the constants and the assumptions
Natural gas at 1,037 BTU per cubic foot, propane at 91,500 BTU per gallon, No. 2 heating oil at 138,500, 3,412.14 BTU per kWh. HSPF2 divided by 3.41214 to estimate seasonal COP. Electric resistance backup at COP 1.0, weighted harmonically. All four assumptions carry the caveat that they vary in the real world.
Wrote a margin rule, and refused to answer where the data will not support one
No winner is declared inside a 10% cost difference, because seasonal price movement is larger than that. Six states are excluded from gas comparisons outright. Declining to answer is the hardest thing to do on a page whose entire purpose is answering, and it is the thing that makes the other answers worth trusting.
Published where the model fails
The methodology page lists the conditions under which the tool is the wrong instrument: time-of-use rates and demand charges, utility heat pump tariffs, partial retrofits where a single whole-house unit is assumed, standing charges on gas bills, and utility-level variation hidden inside a state average. A reader on a time-of-use rate finds that out from the site rather than from their first winter bill.
Ran a corrections log and a research log
Errors are published with a date, the old numbers and the new ones. Open questions we have not resolved are listed rather than quietly dropped. Both of those are ordinary practice for anyone publishing numbers professionally and close to absent in this category.
Released the computed figures
The computed figures are published under CC BY 4.0, so anyone can reuse them with attribution. That is a link-earning mechanism that costs nothing and requires no outreach: the natural way to cite a number you reused is to link to where it came from.
The correction, and what it cost
Published 31 August 2026
The pipeline had been pulling the most recent monthly EIA figures. For natural gas those fell in May and June, when residential demand collapses and fixed charges dominate the apparent per-unit price. Gas therefore looked more expensive than it is over a winter, and heat pumps looked better than they are.
Fixing it, by moving all states to a consistent December-to-February average, cut the number of states where a heat pump wins on running cost under cold-climate conditions from nineteen to four.
Two other corrections were published the same day: a claim about the share of American homes heated by gas was wrong and was replaced with the correct split, and an analysis that had dismissed a homeowner's documented January usage as impossible turned out to be wrong once the home's actual construction was accounted for. In that case the homeowner was right and the site was not.
The first of those made the site's own headline finding substantially weaker. It went up anyway, with both numbers side by side, because the alternative is a site whose figures move without explanation. The site's stated policy is that a correction is published with the old and new numbers together rather than quietly edited out.
We put this in a case study for a plain commercial reason. Any agency can show a chart going up. Very few can point at a live log of what they got wrong, and the ones that can are the ones worth hiring.
Why this is an SEO case study and not an engineering one
None of the above looks like conventional SEO, and all of it is. Search engines are explicit that they are trying to identify content produced with genuine expertise and evident care, and answer engines retrieve and quote passages that state a specific, checkable fact. A published constant, a stated window with a reason, and a dated correction are all exactly that shape.
| What we built | Why it is search work |
|---|---|
| Live API pricing across 51 states | A number no competing page holds, refreshed on a schedule rather than going stale |
| Calculator-first page design | Matches the format the results page shows Google already rewards for the query |
| Published constants and assumptions | Specific, liftable, attributable passages, which is what answer engines quote |
| Stated failure conditions | Answers the follow-up query instead of losing the reader to a competitor |
| Corrections and research logs | Evidence the figures are maintained, for readers and for the systems ranking them |
| CC BY 4.0 on computed figures | Reuse produces citations and links without an outreach campaign |
What we are not claiming
Being precise about the limits
- No traffic or ranking figures appear here. We have not published them, so treat this as a case study about method rather than about outcomes. Everything asserted above is visible on the client site and can be checked.
- Not all of this is ours. The engineering judgement, the constants and the final editorial calls belong to the client. We did the search work and the publishing discipline around it.
- Original data is not always available. This engagement had a public API sitting unused in the sector. Many do not, and the honest version of this method in those cases is specificity and first-hand experience rather than a dataset.
- The corrections were real errors. We are presenting the handling of them as good practice, not the errors themselves as an achievement.
- One engagement is not a pattern. It is one site in one category, described accurately.
Questions people actually ask
Why is this client named when the others are anonymized?
Naming a client is their decision and not ours. Ash Banerjee gave permission for this write-up because the site itself is public and the method is the point of it. The other three engagements on this site are anonymized because those clients preferred it, which is the default.
Does publishing your errors not cost you credibility?
Not in our experience, and the incentives run the other way. A dated correction with both numbers visible tells a reader that someone is checking. Silently changing a figure tells them nothing unless they noticed the original, in which case it tells them something considerably worse.
Could a competitor just copy this?
They can copy the page in an afternoon. Rebuilding a maintained pipeline against a government API, deciding on a seasonal window and defending it, and running a corrections log for a year is a different proposition, and it is the part that would have to be sustained rather than shipped once.
Do you only work on sites with data like this?
No. A public dataset nobody in the sector has published is the strongest starting position, not the only one. Where there is none, the differentiator is the specificity that comes from having actually done the work. The service page sets out both.
How this was made: written from our own engagement records and checked line by line against the client's live methodology and corrections pages, both linked below. Drafted with AI assistance, then reviewed and approved by the named author before publication. Our editorial standards.
Sources
- heatpumpmath.com. Methodology: data sources, constants, assumptions and where the model fails.
- heatpumpmath.com. Corrections log, including the 31 August 2026 heating-season correction.
- U.S. Energy Information Administration. Open Data API v2.
- Creative Commons. Attribution 4.0 International (CC BY 4.0).
The method behind it
- SEO for founders — the service this engagement is an example of
- AI search visibility — the answer-engine half of the same work
- How to get cited by ChatGPT, Perplexity and Gemini
- Our own editorial standards, and how we handle corrections
- The other three engagements
- AEO for companies, the service this demonstrates
If there is data in your sector nobody has published
That is usually the fastest route to pages a competitor cannot copy. The first call works out whether it exists in your case, and whether the constraint on your site is technical, editorial or competitive.
Book a 30 minute call