How I'd grow TypingMind

TypingMind is a chat app for AI models. You bring your own API keys from OpenAI, Anthropic, Google, OpenRouter and about 20 other providers, and chat with all of them in one place. It’s built by Tony Dinh, who has been building it in public on X since he bought the domain in March 2023.

It makes money two ways. People pay once for a licence ($39, “once + pay per API use”, next to “$20/mo” for ChatGPT Plus or Claude Pro on its own pricing page). Companies pay monthly for TypingMind Teams, from $83 a month, to run a chat workspace for their staff.

So TypingMind’s buyers are people who pay for AI by the token, or are about to. That makes the searches it should win easy to name: what the API costs, how to get a key, and which model to use.

Two thirds of TypingMind’s search traffic is brand search

Ahrefs estimates typingmind.com and its subdomains get about 8,600 visits a month from Google, with a Domain Rating of 70. This map shows where they land.

About 68% of the visits are brand searches. That’s a good sign: people know TypingMind, and Google treats it as a known brand, which makes the non-brand pages below easier to rank. The provider guides look better than they are. Most of that traffic is one page: the Groq guide ranks on page one for “console.groq”, for people looking for Groq’s own console.

The surprise is the bottom of the map. TypingMind has 7,323 pages that price every model at every provider, and they get 166 visits a month between them. One page on BenchLM, a site with a lower Domain Rating, gets about 7,300 on its own.

That gap, 166 visits against 7,287, is the largest in this teardown, and the other three are like it. TypingMind has already built the pages. They’re titled, split or hosted in a way that keeps them out of the results.

The four changes that bring most of the gain

1. Retitle the price pages for the search. “claude api pricing” gets 7,400 US searches a month at a difficulty of 3, and TypingMind has the data.
2. Retitle the API key guides the same way. “anthropic api key” and “claude api key” get 35,000 US searches a month between them, and TypingMind has two pages that answer them.
3. Turn the side-by-side model tests into comparison pages. “claude vs chatgpt” gets 35,000 a month at a difficulty of 8, and TypingMind’s X account already runs these tests.
4. Bring back the Teams comparison pages. Their old addresses return a 404, and AI answers about team chat cite LibreChat instead.

Then three technical fixes (a sitemap, links in the homepage HTML and redirects for the old docs) let Google and AI crawlers find all of it.

1. API pricing: the data is there, the title isn’t

Every price page sits under TypingMind Teams’ site, custom.typingmind.com, in a free tool called the LLM cost estimator. There’s one page per provider (100 of them) and one per model at each provider (7,222).

These are the searches for API prices, with Ahrefs’ US volumes. The ones with low difficulty are wide open.

BenchLM’s Claude page shows what wins. Its title is the search, the month and the price range per 1M tokens, every Claude model sits in one table near the top, and Google’s AI Overview for “claude api pricing” cites it. One of its sections is called “Subscription or API?”, which is TypingMind’s whole pitch.

TypingMind’s Anthropic page has the same prices for 32 models. Its title is about the team plan.

32 pages for one model

The second problem is that each model is listed once per provider that sells it. Claude Opus 5.5 has 32 pages. 25 of them show Anthropic’s own price, so Google sees 25 near-identical pages and has to pick one. For “claude opus pricing” it picked the page for Opus 4.6, an older model, and put it tenth.

One page per model, with every provider in a table, is also a better page. Nobody else in the results shows that Bedrock’s regional endpoints cost 10% more, or which gateways charge Anthropic’s list price.

The fix:

1. Title each provider page for the search: “Claude API pricing”, “OpenAI API pricing”, “Gemini API pricing”. Put the price table first in the HTML, with the month in the title.
2. Merge each model’s provider pages into one page with a row per provider, and 301 the rest to it. That turns 7,222 model pages into roughly the number of models, each with something the others don’t have.
3. Add a “Subscription or API?” section to the Claude and OpenAI pages that uses the calculator. TypingMind already compares “$39 once + pay per API use” with “$20/mo” on its own pricing page. Then give that comparison a page of its own for “is chatgpt plus worth it” (3,700 searches, difficulty 0).
4. Move the tool to www.typingmind.com, where the links are. The main site has about 2,650 referring domains and the Teams site about 800 (Ahrefs).

To confirm it worked, check in Search Console that the Claude page gets impressions for “claude api pricing” within two weeks, and that only one Opus 5.5 address shows up for “claude opus pricing”.

2. API keys: the moment someone becomes a TypingMind buyer

Someone searching for an Anthropic API key is about to pay for Claude by the token. The next thing they need is somewhere to use the key, which is what TypingMind sells.

Anthropic’s own pages hold the top spots, as they should. Below them the results are open: a YouTube video, a Reddit thread and two small guides.

TypingMind has two pages that answer it. The docs page “Anthropic Claude” has the four steps (create an account, add credit, create the key, paste it into TypingMind) in 241 words and no meta description. The guide “How to use Anthropic API Key for AI chat” on the main site has the same steps in about 1,500 words. Both are titled for the second step, after the searcher has a key, and they compete with each other.

The fix: make the main-site guide the one page per provider, titled with the search (“How to get a Claude API key”), and 301 the old guide address to it. Open it with one sentence that answers the question on its own: “To get a Claude API key, create an account at platform.claude.com, add credit under Billing, then create a key under Settings, API keys.” Then the steps, what a message costs (linked to the price page) and how to chat with the key. Keep the docs page for setting up TypingMind, and link it to the guide for getting the key.

The same template covers every provider TypingMind supports. OpenRouter is worth its own page for free models: TypingMind already has a guide for each OpenRouter model, so a list of the free ones is a filter over pages it has.

3. Model comparisons: the side-by-side tests live only on X

TypingMind’s best feature for a comparison is in the product: send one prompt to several models and see the answers side by side. The company’s X account uses it in public. “Kimi K3.1 Pro vs. Claude Opus 5.5: Drawing clouds in the sky” got 13,175 views on 26 October, the account’s best post in the last two weeks.

Those tests have no web page. The full chats are shared on cloud.typingmind.com/share/, which robots.txt blocks, so Google never sees them. People search for these comparisons in large numbers, at very low difficulty.

For “claude vs chatgpt”, Reddit, Hackernoon and Zapier rank in the top ten with written comparisons. TypingMind can publish what none of them have: the same prompts sent to both models at once, the full answers, and what each answer cost at API prices from its own price data. And it can end with the answer the searcher didn’t know about, which is to use both with your own keys and stop paying two subscriptions.

The fix: the playbook for any post that does well. Publish the long version on the web at /compare/claude-vs-chatgpt, post a short version that links to it, and add the link in a reply under the original post. Start with the pairs above, and add a page when a new pair starts getting searches (Ahrefs already shows 450 a month for “opus 5.5 vs astra”).

Tony’s own account matters here. The company account has about 5,200 followers and Tony has about 207,000. His post about the first $10,000 in four days got 893,000 views. When a comparison page or the price report below goes live, a post from him is the fastest way to get it read and linked.

4. Teams: the comparison pages that used to rank

TypingMind Teams is the subscription line, and it’s where AI answers leave TypingMind out completely. For two questions about chat for teams, TypingMind was cited or named in none of six answers. LibreChat was in five.

The docs used to have the pages these buyers want. Their addresses still have links pointing at them, and now return a 404:

The searches behind them are team plan prices, and TypingMind’s price data covers half of each answer.

The fix: rebuild the two comparison pages on the main site, at /teams/vs-chatgpt-business and /teams/vs-claude-team, and 301 the old docs addresses to them. Lead with the cost for a team of 20 at both, worked out with the calculator, then the parts a buyer checks: which models, where the data is stored, single sign-on. The Teams pricing page already has the numbers for TypingMind’s side ($83 to $249 a month, and enterprise from $18,700 a year).

Links: the integration pages that list everyone but TypingMind

Links come to sites that others have a reason to cite. TypingMind connects to more than 20 AI providers, and the providers’ own pages that list compatible apps don’t include it. These are pages that link to two or more of LibreChat, Open WebUI, Msty and ChatHub. None of them mentions TypingMind.

OpenRouter’s page is the one to start with. Each app it lists gets its own page there, with setup steps and a link back. TypingMind would need a short request and the OpenRouter guide it already has.

The lists that rank for “chatgpt alternative” (23,000 US searches a month) leave TypingMind out too. Neither GoodDay’s list (second) nor Zapier’s (third) mentions it.

A price report only TypingMind can publish

TypingMind tracks 7,323 model listings across 120 providers. With a date and a finding, that’s a dataset other sites can quote. From one model alone: Claude Opus 5.5 is sold by 32 providers, 25 of them charge Anthropic’s list price, and Amazon Bedrock’s regional endpoints charge 10% more than its global one.

A monthly report would work as the headline: “Claude Opus 5.5 is sold by 32 providers in November 2026, and 25 charge the same price.” The Teams side has a second dataset, with more than a million messages a month across 873 company workspaces (from TypingMind’s About page). “Which models companies actually use”, counted from those messages, is a number nobody else can publish.

An agent can work through the list above and the writers who cover AI pricing, with the TypingMind team approving each email. Ian’s cold outbound method has the steps.

AI answers cite TypingMind as an app and skip its price and key pages

TypingMind does well when someone asks for a chat app. It’s cited or named in 13 of 30 answers to ten buyer questions, against 18 for LibreChat. The gaps are the pricing, API key and teams questions, the same three as in Google.

AI crawlers can read TypingMind’s pages. The homepage, a guide, a skill page and the Anthropic price page return the same full page to a browser and to the GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot user agents. The docs, on Mintlify, send OAI-SearchBot, PerplexityBot and ChatGPT-User a clean Markdown copy. So the gaps are in what the pages say and where they link.

AI answers quote sections that make sense on their own, so each new page should open every section with a sentence that answers the question and names the subject. The About page needs one sentence that names the company, the category and who it’s for: “TypingMind is an AI chat app for people and teams who use ChatGPT, Claude, Gemini and 20 other providers with their own API keys.”

The pricing and API key questions went to Anthropic, BenchLM and Merge, which have a page titled with the question. The fixes in sections 1 and 2 are the AI fix too. For the teams questions, the Teams comparison pages and the integration lists above are what’s missing.

Three technical fixes, closest to revenue first

1. The main site has no sitemap

www.typingmind.com has no sitemap, and robots.txt doesn’t point to one. That site holds the 2,772 model guides, 72,641 skill pages and 8,207 MCP server pages. Google can only find them through the menu and the directories’ numbered pages. The skills directory stops at page 200, which reaches 10,200 of the 72,641 skills, and any later page number shows page 200 again.

www.typingmind.com/robots.txt
  User-agent: *
  Disallow: /share/
  Disallow: /plugins/
  Disallow: /characters/
  Disallow: /artifacts/

www.typingmind.com/sitemap.xml
  404 Not Found

www.typingmind.com/robots.txt
  User-agent: *
  Disallow: /share/
  …
  Sitemap: https://www.typingmind.com/sitemap.xml

www.typingmind.com/sitemap.xml
  a sitemap index: guides, price
  pages, comparisons, main pages,
  then the skills and MCP servers
  worth indexing

The fix: a sitemap index with the guides, the price pages and the comparison pages first. For skills, list the ones with a reason to rank. The UI UX Pro Max skill page ranks ninth for its name (3,700 US searches), behind GitHub, Reddit and five other sites about the same skill, so 72,641 pages built from GitHub repos won’t rank just by being listed. Pages that add what only TypingMind has (which models run the skill, and a button to try it with your own key) will.

To confirm it worked, submit it in Search Console and check the indexed count for the guides a week later.

2. The homepage has no links in its HTML

The homepage gets 4,584 of the main site’s Google visits, and its HTML links to nothing. The menu and footer appear only after JavaScript runs. Google renders JavaScript, but a crawler or agent that reads the HTML as served gets no links at all. The homepage also has no canonical tag, and /home serves the same page under the same title.

GET www.typingmind.com/
200 OK, 202 KB of HTML
about 2,400 words

links to other pages: 0
canonical tag: none

The menu and footer (pricing,
skills, MCP, docs, Teams) are
added by JavaScript later.

GET www.typingmind.com/
200 OK

<link rel="canonical"
  href="https://www.typingmind.com/">

<nav> and <footer> in the HTML,
with plain <a href> links to
/buy, /skills, /mcp, /guide,
the price pages and Teams

/home → 301 → /

The fix: render the menu and footer on the server as plain links, add a self canonical, and 301 /home to the homepage. The pricing pages for both the licence and Teams need a canonical too, because every footer link to the Teams site adds ?ref=typingmind. To confirm, request the homepage with curl and count the links.

3. 342 old docs addresses return a 404

When the docs moved to their current structure, the old addresses weren’t redirected. Of 585 docs addresses the Wayback Machine has and today’s sitemap doesn’t, 342 return a 404. They include the Teams comparisons above, “get-openai-api-key” and the old setup page for every provider. GitHub, the Cloudron forum and Glarity still link to some of them.

The fix: map each old address to its new page and add the redirects in Mintlify’s redirects setting. To confirm, run the old list through a crawl again and check for zero 404s.

One domain instead of five

TypingMind runs on www, docs, custom (Teams), feedback (changelog) and typingmind.app. The links are split the same way.

The price pages are the move that matters. Under www they’d share the main site’s links; on the Teams site they get a third as many.

Ship it all today, then let it settle

None of this needs a four-week sprint. Each fix is a separate job, so coding agents can work on all of them at once:

1. 1

   Price pages

   Retitle the provider pages for "Claude API pricing" and the rest, put the table first, merge each model's provider pages into one with a row per provider, and move the tool to www.typingmind.com/pricing.

2. 2

   Subscription or API

   A section on the Claude and OpenAI price pages, and a page for "is chatgpt plus worth it", built on the calculator.

3. 3

   API key guides

   One guide per provider, titled "How to get a <provider> API key", opening with a one-sentence answer. 301 the old guide addresses, and link the docs pages to them.

4. 4

   Comparison pages

   Claude vs ChatGPT, DeepSeek vs ChatGPT, Gemini vs ChatGPT, Grok vs ChatGPT, Claude vs Gemini and Opus vs Sonnet, each with the side-by-side answers and their cost.

5. 5

   Teams comparisons

   TypingMind Teams vs ChatGPT Business and vs Claude Team, with the cost for 20 seats, and 301s from the old docs addresses.

6. 6

   Sitemap and homepage

   A sitemap index on www, the menu and footer as plain links in the HTML, canonicals on the homepage and both pricing pages, and /home redirected.

7. 7

   Old docs redirects

   Map the 342 old docs addresses to their new pages in Mintlify.

8. 8

   llms.txt and About

   Links to the price pages, guides and comparisons in llms.txt, and an About page that opens with what TypingMind is and who it's for.

9. 9

   Integration listings

   Request a listing on OpenRouter's Works with OpenRouter page, and draft a short note to each list in the links table and to the ChatGPT alternatives lists.

Someone at TypingMind still needs to check the comparison pages and approve each outreach email. The titles, redirects and sitemap need only a quick review.

Then I’d spend 30 days checking the results:

1. Day 0

   Day 0: Ask Google to recrawl

   Submit every new and changed URL to Index My Site, so Google recrawls them now instead of on its own schedule.

2. Day 3

   Day 3: Check indexing

   Confirm the merged price pages, the key guides and the comparison pages are indexed, and the old addresses redirect.

3. Day 7

   Day 7: First impressions

   Look for impressions on "claude api pricing", "anthropic api key" and "claude vs chatgpt" in Search Console.

4. Day 14

   Day 14: One page per model

   Check that a single Opus 5.5 address shows for "claude opus pricing", and that the redirected copies have dropped out.

5. Day 30

   Day 30: Licences and trials

   Compare licence sales and Teams trials that started from search with the month before.

6. Day 30

   Day 30: AI answers

   Ask the same ten questions with ai_answers, for TypingMind and LibreChat, and compare.

Get the changes seen quickly

Most of these fixes change pages Google already knows under a different title or address. Google won’t recrawl 7,000 price pages on your schedule. I’d submit each new and changed URL to Index My Site the day it ships, so Google is asked to crawl it straight away. That matters most for the merged price pages, where the redirects only count once Google has seen them.

How I ran this

Everything in this teardown came from SEO Skill, run by an agent. You can run the same audit on your own site.

1. Connect the hosted MCP server. It’s free. You sign in with Google, and there’s no password and nothing to install. In Claude Code:

claude mcp add --transport http seoskill https://seoskill.dev/mcp

Codex, Cursor and Claude on the web work too. The MCP guide has the steps for each one.

2. Add a research key. On your account page, save an Ahrefs or DataForSEO key. The key gives your agent search volumes, rankings and results pages, like the tables above.

3. Install the CLI for crawling. The crawl runs on your machine, so it can read any site:

npm i -g seo
seo login

TypingMind’s main site draws its menu with JavaScript, so a plain crawl stops at the homepage. Run seo crawl with --rendering on for sites like that.

4. Ask your agent. This is close to the prompt I used:

Audit example.com and every subdomain we use.
Crawl each one with seo crawl, with rendering on
if the homepage has no links in its HTML.
Count the pages each template makes, and check
their titles against the search each should win.
Use seo_research to find the searches our buyers
make, who ranks for them and whether we can compete.
Ask ai_answers ten buyer questions for us and for
our biggest competitor.
Rank every fix by how close it sits to revenue.

On your own site, the hosted server adds the part I couldn’t see from the outside: your real Search Console clicks and queries. With those, you get measured numbers where this post has estimates.

If you’re at TypingMind, I’m happy to go through any of this. And if you’d like the same teardown for your site, SEO Skill will run it for you.