DEV Community: exploreridder

Do You Need an llms.txt File? What It Does, What It Does Not, and What to Fix First

exploreridder — Mon, 26 Oct 2026 02:43:40 +0000

In this article

What llms.txt is
What the evidence says
What actually decides whether AI can read you
Structure is not the same as being worth citing
What I would do, in order

We publish an llms.txt file on our own site. It is a short markdown file that describes Geology in two sentences and lists the pages we would most like a language model to read. I am not going to pretend it is doing much. When people ask me whether they need one, my answer is "sure, if you have ten minutes, but please fix three other things first."

"llms.txt" is searched about 5,400 times a month in the US, and plugins now generate the file in one click. Here is what it is, what the evidence says, and what I would check before you spend any more time on it.

What llms.txt Is

llms.txt is a proposed convention, documented at llmstxt.org: a plain markdown file at /llms.txt with a short description of the site and a curated list of links, sometimes with a fuller /llms-full.txt holding the content itself. It is not a standard from any search engine or AI company, and it does not grant or block access to anything. That job still belongs to robots.txt.

What the evidence says

In June 2025 Google's John Mueller wrote on Bluesky: "FWIW no AI system currently uses llms.txt." He added that consumer chatbots fetch pages for training and grounding, "but none of them fetch the llms.txt file," which he said is obvious from server logs (Search Engine Roundtable).

You can check this on your own site in a few minutes. Search your access logs for requests to /llms.txt and note which user agents made them. If the only visitors are SEO tools and curious developers, you have your answer.

None of that makes the file harmful. It is cheap to publish, and some developer tools and coding assistants do read documentation in this format, so for a software company with public docs it can be a real convenience. For most brands trying to be recommended in ChatGPT, Perplexity or Google's AI answers, it is simply not where the gains are.

What actually decides whether AI can read you

File or setting	What it controls	Why it matters more than llms.txt
robots.txt	Which crawlers may fetch which paths	A blocked AI crawler cannot read the page at all
Server-rendered HTML	What a fetcher sees without running JavaScript	Many AI fetchers do not run JavaScript
XML sitemap	Which URLs exist and when they changed	Search engines that feed AI answers use it for discovery
Page status and canonicals	Which version of a page counts	Broken or conflicting signals keep pages out of indexes

We keep a list of the AI crawlers and what each one does in robots.txt for every AI crawler. The mistake we run into is rarely a missing llms.txt. It is a robots.txt rule (how to set one for AI crawlers), or a security or CDN setting, that blocks GPTBot, OAI-SearchBot or PerplexityBot without anyone having decided to.

Structure is not the same as being worth citing

A lot of the llms.txt enthusiasm comes from a broader belief that tidy, machine-friendly structure is what gets a page cited. We tested a closely related claim. In our study of 3,352 AI citations across B2B SaaS buying queries, schema markup showed almost no relationship with how often a page was cited (r = 0.103), and pages with FAQ blocks were cited slightly less often than pages without them. What did line up with citations was depth of coverage and outbound links to good sources (GEO for B2B SaaS).

Structure helps machines parse a page. It does not make the page worth quoting.

What I would do, in order

Check robots.txt for AI crawler rules nobody chose. Our free robots.txt tester shows them crawler by crawler.
Fetch your key pages without JavaScript and confirm that prices, product names and your main claims are in the HTML.
Keep the sitemap current (our free sitemap checker flags broken entries) and submitted in Google Search Console and Bing Webmaster Tools. Bing's index has long been one of the sources behind ChatGPT's web search, and it is the one most teams forget.
Make the pages worth citing: cover the buyer's question completely, put the facts early, and link your sources.
Then, if you like, add llms.txt. Ours took a few lines. Just do not report it as an AI visibility win until your logs show something reading it. If llms.txt becomes a real standard, the sites that did steps one to four will benefit from it most anyway.

Customer service outsourcing for small online stores: the best options and what they cost

exploreridder — Tue, 06 Oct 2026 10:26:19 +0000

A small online store should outsource support once the founder or an ops hire is spending hours a day in the inbox, or when evenings and weekends go unanswered. Start with a small pilot on email and chat. Atidiv is the best option for a small store that plans to grow, because one team covers pre-sales, orders and returns and can scale 200 to 300% in two weeks. TalentPop is the lowest-cost start on Shopify.

When does a small store need outside support?

The founder answers tickets every day.
Nights and weekends wait until Monday.
Order edits, returns and refunds pile up behind the questions.
A promotion or a holiday doubles the inbox, and there is no one to call in.

Model	How it works	Suits
Per agent hour	You pay for the hours worked	Steady volume, part-time cover
Dedicated agent	One or more agents work only on your store	Stores with daily volume
Month-to-month	No long-term contract	Brands testing outsourcing
Pilot first	A few agents on a slice of real tickets before team size is set	Anyone choosing between providers

The best options for small stores

1.Atidiv

Atidiv staffs a store's whole order cycle as one queue: pre-sales questions, purchase problems such as failed payments and discount codes, and post-purchase order status, returns and refunds. It works in Shopify, WooCommerce, BigCommerce and Magento, in Zendesk, Freshdesk or Intercom, and answers email in under two hours. A small store can start with a few agents and grow the team 200 to 300% inside two weeks when a peak arrives. Its average QA score is 98%.

2. TalentPop

Dedicated ecommerce support from $8 an hour, Shopify-trained, with no long-term contracts.

3. Influx

24/7 support with pay-as-you-go, month-to-month pricing and no setup fees.

4. SupportYourApp

24/7 support across chat, email, phone, and social in 60+ languages.

5. Hire Horatio

24/7 customer service with ecommerce support for inquiries, order management, and fraud prevention.

Start small

Hand over one channel and the three most common contact reasons first, usually order status, returns, and discount codes. Read the first two weeks of transcripts yourself. If they read like your brand, add the next channel.

What would your store hand over first?

Payback on a factory heat retrofit, worked through

exploreridder — Mon, 28 Sep 2026 13:30:29 +0000

Simple payback is capital cost divided by annual savings. The formula is trivial. Everything interesting is in the two inputs, and most retrofit proposals get the second one wrong.

We work a complete example below on published Indian prices, then list the four items a simple payback leaves out.

Step 1: define the duty, not the plant

Take one load. A plant-wide number cannot be verified and cannot be approved.

For this example: a hot water duty of 500 kW, running 6,000 hours a year, at an output temperature of 70°C. Heat is supplied today by a gas-fired boiler.

Annual useful heat is 500 kW multiplied by 6,000 hours, which is 3,000,000 kilowatt-hours of heat.

Step 2: price the heat you buy today

A gas boiler runs at about 85% efficiency. PNG for industrial users in Gujarat was ₹68 per standard cubic metre in June 2026. Gross calorific value is near 9,500 kcal per standard cubic metre.

Convert. One kilowatt-hour is 860 kcal. A standard cubic metre delivers 9,500 kcal gross, and 8,075 kcal after the 85% boiler efficiency. That is about 9.39 useful kilowatt-hours per standard cubic metre.

So delivered heat costs ₹68 divided by 9.39, which is about ₹7.24 per useful kilowatt-hour.

Annual fuel cost for the duty is 3,000,000 multiplied by ₹7.24, which is about ₹217 lakh. The fuel-by-fuel comparison behind this step is set out at heat pump against a boiler.

Step 3: price the same heat on a heat pump

The duty is 70°C output from a warm source, so a COP near 3.5 is a reasonable working figure. Confirm it with the supplier at your actual source and output temperatures before you build a case on it.

Electrical input is 3,000,000 divided by 3.5, which is about 857,000 units.

On the Maharashtra HT-I energy charge of ₹8.44 a kVAh for FY2026-27, that is about ₹72 lakh.

Add the demand charge. Maharashtra HT-I is ₹650 per kVA a month. A 500 kW thermal duty at COP 3.5 draws about 143 kW, so roughly 160 kVA at a 0.9 power factor. That is about ₹12.5 lakh a year.

Total running cost is about ₹85 lakh, against ₹217 lakh on gas. The annual saving is about ₹132 lakh.

Step 4: divide

Simple payback is capital cost divided by ₹132 lakh. If the installed project costs ₹2 crore, payback is about 18 months. If it costs ₹4 crore, payback is about three years.

Capital cost is the input we have deliberately left open, because it moves with capacity, output temperature, heat source and tie-in scope. Ask for it as an installed figure and not as a machine price.

The four things simple payback leaves out#
The sanctioned load increase. Moving heat from fuel to electricity raises connected load. In Maharashtra, an HT application fee is ₹2,660, and a service connection at 11 kV to 1,000 kVA is ₹2,38,110 overhead. A metering cubicle with CT and PT at 11 kV is ₹91,500.

The security deposit. Maharashtra sets security at twice the average billing cycle and recalculates it annually. A heat pump that adds ₹6 lakh to the monthly bill triggers roughly ₹12 lakh of additional deposit. Gujarat is harsher, computing the deposit as if it were a new service for the entire load.

Accelerated depreciation. The Income Tax rules allow 40% first-year depreciation on qualifying energy saving equipment, and heat pumps are named in Appendix I, item 8(ix). That pulls the cash benefit forward and is the one omission that works in the project's favour.

Power factor. Maharashtra bills HT-I per kVAh and not per kWh. A large compressor pulls the power factor down, so an APFC panel sized alongside the machine can pay for part of the installation.

What the hurdle rate should be

Indian buyers commonly state a preference for one to three years. A World Bank study across 224 Indian small and medium units found realised payback averaging 44 months.

Treat the stated preference as a screening filter and not as a hurdle rate. A project at 30 months is normal for this class of equipment.

Before accepting a supplier's saving figure, check it by the method Tetra Heat Pump's case studies apply to its own four projects.

Where the calculation breaks

The whole result turns on the COP, and the COP turns on the output temperature. At a 60 to 70°C duty, the numbers above hold. At 120°C, measured Indian performance runs nearer 2.0 to 2.26, and the saving falls with it. At COP 2.5 against today's gas price it is still around 53%.

One case reverses. Against delivered coal at ₹6,000 a tonne, heat costs about ₹1.94 a useful kilowatt hour, which a heat pump does not beat on grid power. A coal plant switches for compliance, air quality, labour and customer audit reasons. A payback model claiming a fuel saving there will not survive review.

Running it on your own numbers

Four inputs drive everything above: load in kilowatts, running hours, fuel price and electricity tariff. A calculator that takes those four and returns the saving is at tetraheat.com.