Why ChatGPT recommends Upstash over Redis for Redis: a guide to generative engine optimization
Josh
Co-founder
On this page
"For most early-stage SaaS products, I'd use Upstash Redis." That's what ChatGPT replied to the question "i need redis for my saas, what should i use" on October 8, 2026. We asked ChatGPT the same question on nine days and it named Upstash first on eight of them.
This is even though Redis, the company behind the database, also sells its own managed service called Redis Cloud, which is mentioned a bit more than Upstash by the assistants, but recommended less than half as often.
We ask eight AI assistants the same 20 Redis questions every day for our State of AI Search report on Redis. From October 1 to 9, 2026, they gave 1,434 answers. Upstash ranked first in 623 of them, Redis Cloud in 275 and Amazon ElastiCache in 182.
Much of what puts Upstash at the top is already fully public on upstash.com. For instance, two blog posts go into great detail comparing the pricing and features of nine different Redis providers, and the assistants cite at least one of these posts in 41% of all of their answers. The pricing page itself also lists the free tier on a single line. All of this can then be read by agents as markdown, and they can get a fully working database set up with just one request.
Earlier this week we looked at why AI recommends Clerk, and the answer was the fastest setup. With Upstash, the answer is the pages. Upstash wrote the comparison of the whole category of products, and that is the page the assistants read before they answer.
How often does AI recommend Upstash?
AI assistants recommend Upstash first in 43% of answers about Redis hosting (623 out of 1,434 answers). In 74% of answers, the AI assistants talk about Upstash in a positive or neutral way. Redis Cloud is ranked first in 19% of answers and Amazon ElastiCache in 13% of answers.
Redis Cloud is mentioned in 80% of answers, while Upstash is only mentioned in 76%. However, 118 of the answers that mention Redis Cloud are negative about it, against 22 for Upstash. Most of the negative ones are about the size of its free tier or its price.
Comparison with other providers (out of 1,434 answers):
Upstash is ahead of Redis Cloud in all eight assistants. OpenCode is the most favorable and ranks Upstash first in 111 of 180 answers. Claude in the chat app is the coolest on Upstash, and it still ranks it first almost twice as often as Redis Cloud:
On average, each assistant answered about 180 questions in this window of time.
What is generative engine optimization?
Generative engine optimization (GEO) is the effort to get AI assistants like ChatGPT, Claude, and Gemini to mention your pages and recommend your product in their answers. Search engine optimization (SEO) is about getting your site to rank highly in a list of links, while GEO is about getting your site mentioned in the actual answer.
The name is taken from a 2023 research paper. The authors of the paper tested various techniques to increase the visibility of a web page in AI answers, and the best techniques improved visibility by up to 40%.
Each provider’s GEO is represented by two figures: how often the provider is ranked #1 in an answer and how often the provider’s site is cited in the answers. Upstash leads in both metrics, with the assistants citing upstash.com in 70% of answers and redis.io in 56% of answers.
Upstash wins the first questions a team asks
Upstash's 43% is an average over very different questions. Split by question, Upstash wins almost every answer about starting small and almost none about uptime or EU hosting:
The answers that put Upstash first give the same few reasons. The word "serverless" is in 577 of those 623 answers, and "free tier" is in 479. The answers say it in plain words:
- "It offers pay-per-request pricing, scale-to-zero, a REST API for runtimes where persistent TCP connections are awkward, and a free tier." (ChatGPT, on "what's the best redis provider")
- "managed, serverless Redis with a generous free tier and no servers to babysit." (Cursor, on the side project question)
- "It's specifically designed for teams that want Redis without infrastructure management." (Claude Code, on the small team question)
These reasons come from the product. Upstash has an HTTP API next to the normal Redis connection, so code on Cloudflare Workers or an edge function can use it. Pages only help when the product fits the question.
Clerk had the same shape. It won the questions a small team asks at the start of a project and lost the enterprise ones. With Upstash, the pages the assistants read on the way to these answers are different, and so is who wrote them.
Two comparison posts are cited in 41% of all answers
A total of 3,556 citations to upstash.com were generated by the assistants across 161 pages. The blog was responsible for 45% of those citations, which is more than the docs.
The majority of the work is done by two posts: one compares prices for nine Redis providers, and the other compares the features of each major provider. At least one of these two posts was cited by the assistants in 583 of the 1,434 answers and each post was cited on all 20 questions.
The most cited Upstash pages:
Clerk’s most cited comparison pages were one against one, like Clerk vs Auth0. Upstash has those too. Its two most cited pages, however, cover the whole category, so they can fit a broad question like “what’s the best redis provider” as well as a narrow one about a rising bill.
The pricing post has this table near the top:
The two posts share four things:
- In one table near the top of the post, they present all nine providers with their respective billing model, entry price, and free tier.
- The prices are real and dated. The pricing post says "I checked all prices below against official pricing pages in June 2026".
- They also mention when Upstash is the wrong choice: “It gets expensive when traffic is both high and steady. At 100 million commands a month, pay-as-you-go runs $200”.
- They also give credit to competitors: "ElastiCache is a good pick for constant, high-throughput caching when you're already on AWS and can buy reserved nodes."
The assistants pass on these caveats. ChatGPT's answer for a small team recommends Upstash and adds: "Per-request pricing can become expensive with sustained high traffic." A page that explicitly notes its own weak spot is safe for an assistant to quote.
Assistants learn about Redis Cloud from Upstash's pages
When an assistant looks up Redis Cloud, it often ends up on Upstash's site. Of the 1,142 answers that mention Redis Cloud, 897 cite upstash.com and 702 cite redis.io.
Note that Redis itself has no cited pages comparing it with Upstash. None of the 361 pages on the redis.io domain cited in our data contain Upstash in the URL or the title of the page. The comparisons of Redis with other products are of ElastiCache and Google Memorystore (several of them are PDF datasheets), but none of them compare Redis with Upstash. In our data there are 46 redis.io PDFs cited by the assistants, and 137 of the 141 answers that cite one of them are from Codex.
And so Redis Cloud's limits reach the reader through Upstash's posts. Redis Cloud's free plan holds up to 30 MB. "30 MB" is in 299 answers, and 283 of those cite upstash.com. The free plan also allows 100 operations per second. That limit is in 82 answers, and all 82 cite upstash.com.
Both numbers are correct, and both are on Redis's own pages. However, the assistant picks up Upstash's wording for these numbers. Upstash's pricing post says Redis Cloud's "free tier is 30 MB, which is only big enough for a demo". Claude wrote that sentence, word for word, in four answers.
The pages an answer cites go together with who it ranks first:
An answer that cites only redis.io ranks Redis Cloud first a third of the time. An answer that cites only upstash.com ranks Upstash first two thirds of the time.
Note that these numbers only show a correlation and cannot prove causation. A citation only tells us that the assistant listed the page. However, the gap remains across providers: when an answer cites a provider’s own site, that provider is listed first in 58% of cases for Upstash, 29% for Redis Cloud and 19% for ElastiCache.
Which pages does each assistant read?
Claude, Claude Code, Cursor and OpenCode mostly cite Upstash's blog. ChatGPT cites the docs and pricing page but almost never the blog. Codex cites all three.
How many of each assistant's 180 answers cite each part of upstash.com:
ChatGPT and Codex go to the vendor's own pages on purpose. Of the 4,175 web searches the assistants ran, 1,225 were limited to one site, and ChatGPT and Codex ran all of them. 177 were limited to upstash.com. This is one from Codex:
ChatGPT cited the two comparison posts in only 2 of its 180 answers, but still ranked Upstash first in 81 of them. It found the same prices and limits on the pricing page and in the docs.
So a fact only in a blog post reaches Claude and the coding agents, mostly missing ChatGPT. A fact only in the docs mostly misses Claude. Upstash prices and limits are stated in all three.
Codex reads the most by far, with 43 citations per answer. 179 of its 180 answers cite at least one PDF, and 85 of them cite a slide deck from a Kafka course that Upstash hosts on its site. All of those questions were about Redis.
Put one free tier number on every page
The free tier limit of Upstash, 500K commands a month, is given in 419 of the 1,434 answers in our data, or 29%. It’s also one line on the pricing page: “Free tier includes 250MB data size and 500K commands per month.”
The data size, however, is a bit less tidy. The plan card on the pricing page says 250 MB, while the plan table further down the page says 256 MB. Answers repeat both numbers. 401 answers say 256 MB and 93 answers say 250 MB.
Another figure is given on another page. The docs FAQ has a question about a free limit of "10.000 requests per day", and the assistants cite that page in 118 answers. 41 answers give 10,000 commands or requests a day as the free tier.
We can’t tell whether those 41 answers read that page or if they just remember the number from their training. However, the number is on the site for them to find, and it disagrees with the pricing page.
The assistants repeat whatever number they landed on. If a site has one number in one sentence, the answers agree with each other. If it has three numbers, the answers carry all three.
A database an agent can get with one request
Upstash serves its pages as markdown to a client that asks for it. The pricing URL returns markdown or HTML depending on the Accept header:
The pricing page is 250,952 bytes of HTML. As markdown, it’s 15,537 bytes, which is 94% smaller. For comparison, redis.io/pricing answers the same request with about 1.9 MB of HTML.
The markdown version of the pricing page starts with a section called "For AI Agents: Free Instant Redis". An agent sends one request and gets a database, with no signup and no API key:
The response took 1.5 seconds when we ran it. It is markdown too, shortened here with the credentials removed:
After the credentials have been outputted, example commands follow which can be run straight away with the agent. Please note that Upstash will delete the database after 3 days if nobody claims it.
This is already being reported back by the assistants. 27 answers now name the URL, 25 of them on the question "where can i get a free redis database". Cursor lists it under "Instant DB (no account)".
Upstash also ships an llms.txt plain-text index of a site for AI tools, which includes a "When to use Upstash" section explaining which of its products best fits each job.
In our article on Clerk, we argued that agents favor products they can complete a task with. Here, an agent can finish the task before a person has created an account.
Where Upstash is never first
Upstash is ranked first in 0 of 72 answers on four of the 20 questions. On two of them a hosting provider is rarely the top pick at all. The answers about an open source alternative put Valkey first, and the answers about Django put Django itself first.
The other two are about uptime and EU hosting. The reasons are different, and only one of them is about pages.
The question of uptime is which of the available managed Redis services offers the highest level of uptime and the most robust persistence. Out of the 72 answers given, 32 cite Upstash’s feature comparison, but only 6 mention Upstash as a provider. In contrast, the Redis Cloud Pro plan promises up to 99.999% uptime, and that number is mentioned in 71 of the 72 answers.
As for the uptime guarantee, Upstash guarantees 99.99% uptime for databases with the Prod Pack add-on or an Enterprise plan. The pricing page currently lists the Prod Pack add-on at $200 per database per month. OpenCode sums it up as "99.99% but only if you pay the $200/mo Prod Pack add-on".
The assistants read the pages and reported them correctly, but Redis Cloud has the stronger number. A blog post can’t change that.
The user wants “Redis hosted in the EU for GDPR”. Upstash has locations in Ireland, Frankfurt and Belgium. This is covered in the Upstash docs page on global databases. 42 out of 72 answers mention Upstash, but none of the Upstash pages cited in these answers answer the question fully in one place.
The assistants put the Upstash part of their answer together from five pages:
- The docs page on global databases, in 25 answers
- The docs FAQ, in 19 answers
- The compliance page in the docs, in 16 answers. It answers "Is Upstash GDPR Compliant?" in three lines.
- The privacy policy PDF, in 16 answers.
- The data processing agreement PDF, in 14 answers
They found a page that lists every supported region for Redis Cloud and cited it in 42 answers.
This question could be answered by a single page that lists the existing EU regions, explains how to keep data inside them and links to the data processing agreement.
The playbook
Upstash’s pages are public, so you can read them and copy the moves for your own category.
- Write the comparison of your whole category. Put every provider in one table near the top of the post, with real prices and the date you checked them.
- Be honest on that page about when you are not the right choice for the customer, and give credit to your competitors where they win.
- Write the one-on-one pages as well, although they are cited less often than the pages against several competitors. But when they are, you are ranked first more often.
- Put your pricing and limits on the pricing page, in the docs and in the blog. This way each assistant will read what they need to read.
- Keep one free tier number, in one sentence, and check old pages for old numbers.
- Serve markdown to agents and ship an
llms.txtfile that specifies when they should use you. - Let agents test the product with one request and no signup.
- For each question you miss, look at the pages the assistant cited for that question. If your page for that question is missing, write it.
- Track where AI ranks you – question by question – to see which of these moves pay off.