What AI assistants say about Circle today, and what I'm thinking we do about it.
Arina, JK, Marcus, I wanted to put everything I've learned about how AI assistants see Circle in one place, along with what I think we could do about it and roughly in what order. I'm sharing it to get your take, not to hand you a plan. You may already be on the same page with a lot of this. But I've been sitting with this research for a week and a half now, and the thing I keep coming back to is how much of the fix is small. Not easy necessarily, but small. A lot of it is a few hours of work, not weeks.
Here's the short version. When someone asks ChatGPT, Claude, Perplexity or Gemini what to build a coaching business on, we come up when they type the word "community," and for almost nothing else. When the buyer names us, the assistant knows what we do and gets it mostly right. What it gets wrong is our prices, which it has from 2024, and everything we shipped in 2026, which it can't find. And when I went looking for where the wrong facts come from, most of them come from us. Our own pages, our own machine-readable description, our own blog posts.
My point isn't that the site is bad. A lot of it is fine and some of it is very good. My point is that we're currently the source of our own worst positioning, and that's the part we can fix fastest. That's why I'm writing to the three of you and not to engineering.
Assistants know what we do. They just don't recommend us unless the buyer says community.
I had Quinn, my AI research assistant, run three passes over the last two weeks. One tested what assistants recommend across 28 things buyers say they want to build and 20 real buying situations, asked both cold and with web search on. One fetched all 215 pages on circle.so the way an AI crawler does, without JavaScript, and gave every page a verdict. And one read the 149 blog posts the first two hadn't. The full tables are in the appendix and they're worth a look, because if we do this you'd be writing to these terms. Here's what I took from it.
We wrote twenty realistic buying situations the way a person actually types them. For example: "I'm a coach with 15 one-on-one clients, I want to move them to weekly group calls and charge monthly, what should I use?" Each one was asked two ways, once without mentioning Circle and once naming Circle and asking whether it can do the job. And each was run in two modes: with the assistant's web search off, so it answers from what it remembers from training, and with web search on, so it answers from the pages it finds. That's four answers per situation, eighty in total, each scored for whether we appeared, where we ranked, what it said we can and can't do, and which sites it cited.
So "first or second in nine of twenty from memory" means that in nine of the twenty unaided runs with search off, Circle was the first or second platform named. "Disappear in nine of twenty" means that in nine of the twenty unaided runs with search on, Circle wasn't named at all, because the pages that rank for a yoga studio or a church are niche tools, and the pages that rank for a webinar that sells a course are roundups we aren't in. Every situation, with all four answers, is in the appendix.
One caveat that matters: all of this was run on Claude, standing in for ChatGPT, Perplexity and Gemini, because we didn't have live access to the others. Claude with web search is a close proxy for what those engines find and cite, not their literal answers. The monthly check I mention at the end would run on all four.
The full grid, all twenty business types, is in the appendix. The six rows I'd point you at are the ones marked Opportunity: the five above, plus the one that doesn't have a name yet, where someone asks an AI to set the whole business up for them. On each of those the assistant already thinks of us first or second and has nothing on our site to point to.
The pages assistants read about us are mostly written by competitors, and our own pages still say community platform.
An assistant reads two kinds of pages about us, other people's and ours, and right now both are working against us.
Start with other people's pages. The pages an assistant pulls up about Circle are mostly written by our competitors and their affiliates. Across the first benchmark, circle.so came up on two of eight buyer questions. The domains that came up most were group.app, schoolmaker.com and mightynetworks.com. Mighty ranks first for "Skool vs Circle" and "Circle alternatives." SchoolMaker owns "Circle pricing." GroupApp owns "Circle review." A buyer researching us reads our competitors describing us. The reason is money. Skool and SchoolMaker pay their affiliates 40 percent recurring, and until June we paid $120 once. We moved to $100 plus 10 to 20 percent recurring in June, but the pages assistants read haven't caught up, because nobody has told the reviewers. That one's on Justine and me, and it's not on your list.
Now our own pages. A crawler can read almost every page on circle.so that matters, and on most of them what it reads is the old story. Beyond the wrong facts above: eight business-type pages share a template whose structured data asks "How does Circle compare to other community platforms?", thirteen pages still say "co-pilot" for a product we renamed (as of Sep 23: /compare and five of the head-to-head compare pages, /platform/creators, /platform/courses, /platform/community-managers, /membership, two blog posts and llms.txt), and not one feature or compare page names anything shipped in 2026. Two things a crawler can't read at all. The help center returns a 403 and a "Verifying you are a human" page to every AI crawler that was tried. And our release notes: circle.so/product-updates itself comes through fine, about 950 words linking 29 releases, but every one of those links goes to a post on community.circle.so that renders through JavaScript, so a crawler gets 11 to 14 words, the title and one line. So every "can Circle do X" question gets answered from what the model remembered from 2024.
Everything in this section is checkable in a minute. Open the page, view the source, and search for the phrase. What's in the source is what an assistant sees. Or paste a release note URL into ChatGPT and ask it what shipped. If a fact in this memo doesn't match what you find, tell me, because the site moves. Two things already changed between the Sep 14 crawl and my re-check on Sep 23: "Content co-pilot" is off the pricing page and off most of the platform pages, and the $49 on the pricing page is now the only one on the core pages.
For the record, the crawl gave 182 pages a verdict: 75 keep, 75 fix, 14 rewrite, 10 merge and 8 retire. Every page's row is in the appendix.
The good news is that a person and an assistant want the same thing from a page. It should say what the product does today, in the words a buyer would use, with a number next to it. That's really the whole idea.
Three buckets of hours-long fixes clear the ground. The new pages in bucket four are where I think the trials come from.
I've sorted everything into five buckets, because I think the effort is the thing that gets lost when a list like this lands on someone's desk. Bugs, then the basic blocking and tackling we should have done already, then mistakes on our own pages that are easy to correct, then the real builds, then the engineering and product work that runs alongside. The first three buckets are hours each. The fourth is days each and carries most of the upside. Every row has a score from one to five. The score is how obvious it is to do now, not how big it is: a five is something I don't think needs any debate, and a three is usually something big that needs sizing or a decision first. Each row says which. And each row's title links to its full write-up in the appendix, which has the specifics down to the exact line to change.
If I had to pick one thing, it'd be rewriting llms.txt, which is about an hour. The first three buckets together are maybe two or three weeks of work between two or three people, and once they're done we've stopped feeding the assistants wrong facts ourselves. The term pages in bucket four, and the fees page in bucket two, are where I think new trials actually come from. Everything before them is clearing the ground so those pages get believed when they go up.
Net-new means a page that doesn't exist today. Five items. These are the ones that take real days, and the ones to push out if the month gets tight.
Fixes to what exists: wrong facts, old positioning, broken markup, pages a crawler can't read. Twenty-three items, mostly hours each.
Ordered by how obvious the payoff is for the time. Hours-long fixes to facts the assistants already repeat come first. Multi-week builds come last. Every row keeps its bucket tag.
Ordered by how directly each item changes what an assistant says about us. Things only a model reads, and pages that flip a live answer, come first. General site hygiene comes last, even where it's quick.
| # | What to do (the title links to the full write-up in the appendix) | Why it matters | Why it's ranked here |
|---|---|---|---|
| 1 | Fix the canonical tag on /mcp and list the page.The problem. Our MCP page is telling search engines it's a duplicate of another page, so it never shows up. Point the canonical at /mcp (it points at /chat today), add it to the sitemap and llms.txt, link api.circle.so/mcp, add "since April 2026" and the two access modes. | One line of markup tells search engines our MCP page is a duplicate of the chat page, which hides the page that proves we're ahead of Kajabi and Mighty on this. | It's minutes. That's the only reason it's this high. The payoff is one page, not the whole site. |
| 2 | Fix the bugs a visitor can see today. The problem. Pages that sell are broken or mislabeled in ways anyone can see. Two "Builder block error" messages on /membership. The /plus/branded-app H1 renders as "Earn more with your own." The Analytics Manager card on /ai carries the Notification Manager description. Six of seven featured creators on /discover are captioned "Productivity expert." A "Launch your course" heading on /platform/podcasters. /compare says 14 competitors in the subhead and 17 in the body. | The membership page is a money page and it's broken in public, and this is probably the cheapest credibility fix we have. | Quick and embarrassing, but it fixes what people see, not what assistants say. So it sits below the fact fixes. |
| 3 | Retire the junk pages and redirect the URLs agents guess. The problem. Old pages and dead URLs are still out there being read and quoted. Lorem ipsum on /academy. A Portuguese sales form under an English "Book a Demo" title at /demo-video. An internal note published on /migration/courses ("This one might be worth discussing internally"). /v3, /affiliate-faqs, the two year-review stubs, the two 2023 Copilot posts. Three copy-paste errors in customer stories. 301s for the eight URLs agents guess (/customers, /case-studies, /compare/kajabi, /knowledge-base, /help, /fees, /about, /changelog all 404). | Old pages are what the models end up quoting, and every URL an agent would guess for our pricing or help returns a 404 today. | A day of cleanup. Old pages are what models quote, so it's worth doing early, but nothing here changes an answer on its own. |
| 4 | Hide the nineteen old event-hub transcripts, from 2021 to 2023, from crawlers. The problem. Years-old event transcripts are the biggest block of old-positioning text on the site. | It's about 190,000 words that say "community" fifty to two hundred times a page, the biggest block of text on the site after the blog. | One line of work that removes the biggest pile of old-positioning text on the site. It's below the pricing and llms.txt fixes because those correct facts assistants repeat. This just stops reinforcing a word. |
| # | What to do (the title links to the full write-up in the appendix) | Why it matters | Why it's ranked here |
|---|---|---|---|
| 5 | Rewrite llms.txt, the file on our site written for AI models, and regenerate llms-full.txt.The problem. The one file we wrote for AI models describes us as the thing we're moving away from. Write it as a brief for the assistant reading it, the way Kajabi's opens "Guidance for large language models, AI agents, and automated assistants": what Circle is, the business types we serve, when to recommend Circle and when not to, prices and fees, the 2026 capabilities with plan gating, the MCP server, Discover, the platform pages, all compare pages, all 25 stories, product updates. And one line saying every page has a markdown version if you ask for it. Regenerate the full file from product, pricing, compare and platform pages first, blog last. | This is the file written for machines, and it says "Circle is community-first, not course-first or marketing-first," which is an hour of writing to fix. | The best hour on the list. One file, written for machines, saying the sentence we most want gone. |
| 6 | Fix the plan comparison table on the pricing page. The problem. An add-on price on the pricing page reads like a plan price to a crawler. Plan availability as text instead of icons, monthly prices next to annual, the add-ons in their own clearly labeled table, a fee example at three revenue levels, product and offer structured data. (A markdown version of the page already exists, see the technical section at the end. "Content co-pilot" was in the shared block on Sep 14; it's off /pricing as of Sep 23 and still on thirteen other pages, which is item 12.) | The only $49 a crawler sees on circle.so/pricing is the "Customizable profile fields" add-on at $49/month in the plan table, and with the column headers lost it reads like a plan price (view the page source and search for $49). | The wrong fact assistants repeat most, and it's hours to fix at the source. Only llms.txt is a better hour. |
| 7 | Fix our profiles on G2, Capterra and GetApp. The problem. The review sites still list old prices and file us in the wrong categories. Capterra and GetApp show "from $49"; GetApp files us under HR software. Add G2 categories for coaching, course, membership and newsletter, not only community. | The models use G2's categories to decide what a product is, so if we're not in coaching and courses we don't exist for those questions. | Cheap, and it changes which questions we're even in the running for. Just below the on-site fixes because we're waiting on G2 and Capterra to process it. |
| 8 | Rewrite the one-line description on every profile we control off circle.so, so they all say the same thing. The problem. Our social and directory bios still describe us as a community home, and each one says it differently. Instagram (@CircleApp) currently says "From run clubs to diabetes support groups, Circle helps creators and brands build homes for their communities, memberships, events, courses, and more." X, LinkedIn, YouTube, TikTok, Facebook, Product Hunt (Circleapp), G2, Capterra, GetApp, Crunchbase, the App Store and Google Play listings, and GitHub each have their own line, and none of them match. Once the one-line description you are already working on exists, put the same sentence on all of them, name the business types, and put circle.so in every bio so a model can tie the handle to the site. | These profiles are among the pages a model reads when it works out what Circle is, and they're cheap to change. Today every one of them teaches community, and a handle that never says circle.so is harder to tie to us than the crypto company is. | A morning of work once the line exists, and it touches a dozen pages that models read. It's just below the review-site fix because the review sites feed the category we're filed under, while these feed the description. |
| 9 | Publish a pricing-and-fees page with the total-cost math. The problem. There's no page that walks through what Circle actually costs, fees included. One page that explains every plan and fee in plain words, with the total cost worked out at $5K, $20K and $40K a month of revenue, next to Skool's and Kajabi's. (This was three pages. The other two, "what shipped in 2026" and "what Circle is," are what items 10 and 5 produce once they're done, so I cut them.) | The hidden-fee story shows up in nearly every answer about our pricing, and nothing on our site corrects it with the math. | The one new page in the first three buckets. It stays above the term pages because the fee myth is in almost every pricing answer. It dropped a few spots once I cut the two pages that duplicated items 5 and 10. |
| 10 | Make product updates something an assistant can actually read. The problem. Everything we shipped this year lives in release posts a crawler can't read. The fix, in order. First, the quick steps: give each release its own page on circle.so with the full release note in the HTML (not a link to the community post), add dated article markup, and add the product-updates page and each release to llms.txt and the sitemap. Second, rewrite the entries one per capability, in the buyer's words, with the plan it's on, the date, and links to the feature page and help article. Third, put a "new in 2026" block on the payments, events, website builder, email hub, workflows, forms and AI pages so the feature pages themselves carry the facts. The longer-term product fix, serving community post bodies as real HTML, is item 26. | It's the only first-party record of what we shipped this year and a crawler can't read it: each release post on community.circle.so gives a crawler 11 to 14 words (checked Sep 23, and the page isn't in llms.txt). | Third on my list, behind only llms.txt and the pricing table. Everything we shipped in 2026 is invisible to assistants, and those are the features that separate us from Kajabi and Skool. The first step is a day or two of web engineering, which is the only reason it isn't first. The two above it are pure writing. The second step, per-release pages and rewritten entries, is two weeks and can follow. |
| 11 | Retitle the customer stories with the business type and the number, and add the six missing stories to llms.txt.The problem. Our best proof is titled "community" and the assistants can't find it. Seasoned Sales ($15K MRR in two months), SPI ($700K+), Roofing and Solar, Talk Nerdy to Me ($37K+ a month off Kajabi) first. | What gets us cited is a named customer with a number, and we have that proof for our weakest business types under titles that say "community." | A morning, and every weak business type gets a citable customer. It's below the fact fixes because proof helps a page get cited, and the pages it would help don't exist yet. |
| # | What to do (the title links to the full write-up in the appendix) | Why it matters | Why it's ranked here |
|---|---|---|---|
| 12 | Correct the wrong Circle facts on pages that already rank. The problem. Our own pages state wrong prices, retired plan names and features we don't have. The Kajabi-alternatives fee range, the Patreon-alternatives math, "read-only" Circle AI, "flat regardless of list size," "join the waitlist," "in 2025," API on all plans, order bumps on the Facebook compare page (we don't have them yet), the two compare pages that say we don't issue certificates (we do), any page that puts white-label on Professional (it's Business and above), every mention of Enterprise now that the plan is called Scale, and "co-pilot" on the thirteen pages that still say it. | These are the exact stale facts the assistants repeated back in the testing, and they're on our own pages, so we're the source. | These pages already rank, so a corrected fact reaches assistants the next time they search. A morning of edits with immediate effect. |
| 13 | Fix the template behind the eight business-type pages under /platform.The problem. The template behind the business-type pages calls us a community platform in the part an engine reads first. The FAQ structured data asks "How does Circle compare to other community platforms?" The closing band is titled "The complete community platform." Two pages answer the migration FAQ with a customer quote about launching a course. | Eight pages say "community platform" in the structured data, which is the part of a page an engine reads first, and one template edit fixes all eight. | One template edit fixes eight pages. Same logic as the pricing block: one change, many pages. |
| 14 | Put our price and a fee table on the Skool, Mighty and Kajabi compare pages. The problem. Our comparison pages don't show our own price. Port the table already on the Bettermode page. Fee math at $5K, $20K and $40K a month. FAQ structured data. | Those three pages carry no Circle price at all, which is part of why SchoolMaker owns the search "Circle pricing" and Mighty ranks first for "Skool vs Circle." | Hours, and it starts taking "Circle pricing" back from SchoolMaker. Below the pricing page fix because these pages are secondary sources for price. |
| 15 | Rewrite Circle's own entry in the "best platforms" blog posts of ours that already rank. The problem. Our ranking blog posts describe Circle in retention words, so the assistant reads them and recommends someone else. Best coaching platforms, community, membership, creator, event and membership-site posts, and the Kajabi-alternatives verdict: selling verbs, the 2026 stack, and we win on conversion. | Every sentence the assistant used about us in the coaching test came from our own top-ranked post, and it still recommended someone else, because the post describes us in retention words. | Our posts already rank for these questions. Rewriting our entry changes an answer without having to earn a ranking. It's below the fact fixes only because each post is one answer. |
| 16 | Update the 56 blog posts that still say "community platform" (23 also carry a stale plan or price). The problem. Fifty-six blog posts still say community platform. Start with the two that carry most of it: the 2024 guide to choosing a community platform (14 mentions) and the creator tech stack post we published on Sep 8 of this year (12). Then the rest in batches of ten; most of them are a "community platform" replacement and a paragraph. Full counts per post: blog-sweep-2026-09-17.csv. | Perplexity and ChatGPT both cite our blog, and fifty-six of our posts are telling them the positioning we're moving away from. | Fifty-six posts is real hours, and each one is a small effect. Batch it and let it tail. It's below the ranking pages because most of these posts don't rank for a buying question. |
| # | What to do (the title links to the full write-up in the appendix) | Why it matters | Why it's ranked here |
|---|---|---|---|
| 17 | Write one page for each thing buyers say they want to run: mastermind, challenge, accelerator and the rest. The problem. For the things buyers say they want to run, there's no page on our site with that word on it. Rewrite mastermind, group coaching and cohort on their existing URLs as platform pages: the buyer's word in the title, the mechanics named (live rooms, leaderboards, check-ins, applications, paywalls), one customer with a price, FAQ structured data. Then the ones we don't have: accelerator, challenge, signature program, paid club, Discord alternative for creators, membership site for podcasters, a webinar that sells a course, high-ticket program. | For six of these the assistant already thinks of us first or second, then searches, finds nothing to cite and drops us, and a single page changes that within weeks. | The highest-upside writing on the list. The assistant already thinks of us for these terms, and one page flips the live answer. It's below the corrections only because it's days per page and should land on a clean site. |
| 18 | Put what we shipped in 2026 on the feature pages themselves, because those are the pages an assistant actually reads. Here's the problem. Our feature pages (events, payments, workflows, email hub, analytics, customization, gamification) are readable by a crawler, and they're what an assistant pulls up when someone asks "can Circle do webinars" or "does Circle have upsells." But none of them mention a single thing from 2026. The only place those features are written down is the release posts, and a crawler can't read those (item 10). So the assistant opens the feature page, finds nothing about upsells or webinars, and answers from what it remembered from 2024: no. The fix is to put each 2026 capability on the page it belongs to, in the buyer's words, with the plan it's on. Events: public RSVP, public webinars, paid tickets, every guest becomes a contact, and "webinar" in the title. Payments: post-checkout upsells, installments, Apple and Google Pay, the fee math. Workflows: sale triggers, and Circle AI writing the workflow from a sentence. Email hub, analytics, customization, gamification: fix the FAQs and name the 30-day challenge. | Item 10 makes the release notes readable. This makes the pages an assistant actually retrieves carry the same facts, because a buyer asking about webinars never lands on a release note. In testing the assistant said we don't have upsells and hedged on webinar registration for exactly this reason: the payments and events pages never mention them. | Feature pages are what buyer questions pull up, so this is where the 2026 facts have to live, and item 10 doesn't do it on its own. A week or two across ten pages, which is why it comes after the fees page. |
| 19 | Rewrite our Skool, Mighty and Kajabi comparison pages so they win the searches competitors own, and add the two we don't have. The problem. Our comparison pages lose to Mighty's for searches about us, and two pages that should exist don't. The pages exist at /compare/vs-skool, /compare/vs-mighty and /compare/vs-kajabi. Today none of them shows our price, the Kajabi one is titled "Community Platform Comparison," all three close on "the complete community platform," and Mighty's page outranks ours for "Skool vs Circle." Rewrite them straight, with the fee math and the feature truth, a "what Circle sells that X can't" paragraph and FAQ markup. Then two pages that don't exist: /compare/circle-alternatives, which Mighty currently owns, and a first-party reviews page, which GroupApp owns. | Compare pages are what assistants cite most, and today Mighty owns "Circle alternatives" and GroupApp owns "Circle review," so a buyer researching us reads our competitors describing us. | Compare pages are the most-cited layer and ours already exist, so this is a rewrite that reaches assistants the next time they search. It's a four because Mighty's pages already hold the top spot for two of these searches, and a rewrite competes rather than fills a vacuum. |
| 20 | Write the coaching page, as the first of the business-type pages. The problem. Coaching is our biggest business type and has no page. A headline in their words about the time ceiling, one sourced dollar line in the hero, their money-makers in order with what Circle AI does for each, one story with a dollar figure in the title, the product list in the new nouns, the trial. | Coaching is named on the homepage and is a 404 at every URL a coach tries, and it's the biggest thing creators sell at 31.6 percent per Kit's 2026 survey. | The biggest business type and a 404 today. It's in the builds tier because it's a full page, and the rest of the family gets modeled on it. |
| 21 | Give the selling features their own pages: sales pages, upsells and funnels. The problem. The things we do for selling, sales pages, upsells, funnels, have no page of their own. Three pages that don't exist: sales pages (live today, AI-built), upsells (live today), and funnels as the real sequence we support: public webinar, RSVP, email sequence, checkout, upsell. | Digital products is the most common thing buyers say they want to sell and we have no page, while the deciding feature, one-click upsell at checkout, is something we've had since early 2026 and nobody can find. | Three pages for the most common thing buyers want to sell. High upside, but they need the one-line descriptions you are already working on, and the product answers, first. |
| 22 | Rebuild the website builder page, and use it as the template for every product page. The problem. The website builder page doesn't mention the AI builder. Lead with prompt-to-page and recreate-from-URL. Show the output, then the editor. | It has zero mentions of AI on it weeks after the AI builder shipped free on every plan, and every other product page would be modeled on it. | It's the template for every other product page, so it comes before the rest of the product pages and after the pages that create new answers. |
| 23 | Write the other four business-type pages: education, creator, community and membership. The problem. Four of the five business types have no page, and the shared pieces still say community platform. On the coaching template, then nav, trust line, title tags and the /platform hub, all written from the one-line descriptions you are already working on. | Nobody in the market has a clean page for each business type, and every vertical title tag still says "Community Platform for." | Last of the builds because they're written once from the one-line descriptions, and each one depends on the coaching page proving the template. |
| # | What to do (the title links to the full write-up in the appendix) | Why it matters | Why it's ranked here |
|---|---|---|---|
| 24 | List our MCP server everywhere agents look, and make circle.so/mcp the page that tells an agent how to connect.The problem. Our MCP server isn't listed where agents look, and our MCP page doesn't tell an agent how to connect from anything but Claude. The listings first: the Claude connector directory and the official MCP registry, two forms; ChatGPT's app directory, same server plus one more review; cursor.directory, self-serve; and Meta's Muse once it opens to partners, which someone in Sheel's group should go and find out about. Then the page we already have, circle.so/mcp: add the one-line install for Claude, ChatGPT, Cursor, Grok, Perplexity and Gemini Enterprise, all of which accept a custom MCP URL today, and say plainly what the server can do, tool by tool. And publish a small agents.json beside llms.txt with a "last verified" date, the business types we serve, the plans and prices, and the MCP URL, so an assistant can get the facts in one call. (I looked at whether we need a page like mightynetworks.com/agents. Ramp, Notion, Stripe and Shopify don't have one; they put this in llms.txt and the MCP docs. So do we.) | The Claude directory is the only listing that reports usage back and suggests connectors inside the chat, which is why it comes first. Grok, Perplexity and Gemini have no listing door yet but each takes a pasted MCP URL, so the MCP page covers them today. Right now, ask for the best MCP for a community business and zero community platforms come back. | A day of product work with no downside. It's in the alongside bucket only because it isn't marketing doing it. |
| 25 | Make the help center readable by crawlers. The problem. The help center blocks AI crawlers entirely. Exempt known crawlers from the Cloudflare challenge, server-render article bodies, publish a sitemap. As an interim, mirror the top hundred articles as plain HTML on circle.so. | help.circle.so returns a 403 and a "Verifying you are a human" page to every AI crawler that was tried, so every "can Circle do X" question is answered from 2024 memory. | This fixes hundreds of capability answers at once, but it's weeks of web engineering. The hundred-article HTML mirror is the five. The full fix is the four. |
| 26 | Make public community posts readable by crawlers, and add a robots file on community.circle.so.The problem. A crawler can't read anything posted inside a Circle community, including our own release notes. This is product work, not marketing. | Our own release notes are unreadable to a crawler and so is every customer's public content, and this is the bigger bet: every business on Circle findable by assistants without the customer doing anything. | The score here isn't impact. On impact this is the biggest thing on the list: it's what would make every business on Circle findable by an assistant without the customer doing anything, and nobody in our category has it. It's a 3 because it's months of product work, nobody has sized it yet, and it isn't in marketing's hands. If product sizes it and it turns out to be smaller than months, it moves to the top. |
| 27 | Fix the German, French and Portuguese pages. The problem. The localized pages tell the old story, and most of them are English under the wrong language tag. Port the new positioning to the German, French and Portuguese homepages, and noindex the 180 URLs that serve English under a foreign language tag. | A third of the sitemap either calls us a community platform or serves English under a German or French tag. | Real but slow, and it doesn't change an English answer this quarter. |
| 28 | Get Circle a Wikipedia article. The problem. On Wikipedia, Circle is a crypto company and we don't exist. There is no article for us at all. I checked on Sep 23: "Circle (company)" is the stablecoin company and a search for circle.so returns nothing, while Kajabi has one. Start the notability case now, with the press coverage and the customer numbers, because it takes months. (On Product Hunt we exist as Circleapp; producthunt.com/products/circle is the crypto company, so ours is harder to find, and worth a look while we're at it.) | ChatGPT takes about half its citations from Wikipedia. When a model looks up "Circle," it finds the crypto company, and we have nothing to point it at. | Slow and mostly out of our control, which is why it's last. But it's the biggest lever on what models remember from training, so it should start now, not in December. |
One thing on ownership. I haven't put a name on any of these on purpose. Let's work out who does what once you've all reacted to the list. The only thing I've left off is the homepage and pricing page redesigns, because Sid, Rudy and I own those end to end.
This calendar is an illustration, not a plan. It shows what thirty working days could look like if we put one thing live every day. Nothing on it is decided or assigned.
Here's something I've been playing with, and I want to be clear it's an idea, not a plan I've decided on. What if we took thirty working days, September 28 to November 6, and put something live on circle.so every single one of them? I've laid out what that could look like below, day by day, with the item numbers in parentheses so everything traces back to a row above. The second column is the people outside marketing, on the days they'd be needed. I haven't asked any of them yet; if we decide to do this, lining them up is on me. I haven't assumed anything about what else is on your plates, and I know you all have a lot on. So take this as me showing you what I'm thinking, and tell me what's realistic.
My point isn't that this calendar is right. You'd move things, and you should. My point is that almost everything on it is a day or less, the bigger pages can run in the background and land on their day, and if it worked, by Friday, November 6 we'd have shipped the whole first three buckets, eleven term pages, the fees page, three selling pages, our three comparison pages rewritten plus two new ones, five business-type pages, and most of the 56 blog posts corrected. That's most of the list in six weeks, and it would land alongside the new homepage and pricing page. If a term page needed a second day, I'd let the blog batch slip, not the page.
A few things I'd suggest, whatever the calendar ends up looking like. The first is facts before pages. The first week is nothing but corrections, so that when the term pages land in week three the assistant reading them doesn't also find a $49 tier two clicks away. The second is that claims wait on the answers. A few product facts are still open and I'm working them separately, and no page would claim certificates, order bumps or a currency count until they're settled. The third is one page of each kind before the family. The coaching page and the website builder page would go live first, so the four of us can react to one of each before the other business types and product pages get modeled on them.
| Day | Date | What goes live | Who else is needed that day |
|---|---|---|---|
| Week 1: bugs and facts. Everything here is hours, and none of it needs a meeting. | |||
| 1 | Mon Sep 28 | llms.txt rewritten and live. (5) The wrong facts on the ranking blog posts corrected: Kajabi-alternatives fee range, Patreon math, "read-only" Circle AI, "flat regardless of list size," "join the waitlist." (12) | |
| 2 | Tue Sep 29 | /mcp canonical fixed, page in the sitemap and llms.txt. (1) The Sep 8 creator tech stack post rewritten in the new positioning. (16) | Product and eng: submit the MCP server to the Claude connector directory and the official MCP registry, then ChatGPT's app directory and cursor.directory. Sheel's group: find the Muse partner path. (24) |
| 3 | Wed Sep 30 | The visible-bug ticket shipped with web eng: /membership errors, branded-app H1, the /ai card, /discover captions, 14 vs 17, the podcasters heading. (2) The 2024 guide to choosing a community platform rewritten, or retired and redirected. (16) | Web eng: ships the visible-bug ticket. (2) |
| 4 | Thu Oct 1 | Review-site profiles: four G2 categories added, Capterra and GetApp prices and the HR category fixed. (7) Customer stories retitled with the noun and the number; the three copy-paste errors fixed. (11) | |
| 5 | Fri Oct 2 | The shared pricing block and /pricing, with web eng. (6) Six missing stories added to llms.txt. The coach customer story commissioned. (11) | Web eng: the pricing table changes. (6) |
| Week 2: the facts that have no page yet. | |||
| 6 | Mon Oct 5 | Pricing-and-fees page live, with the fee math. (9) Our entry in "best coaching platforms" rewritten in selling verbs. (15) | Justine: affiliate content brief, paired with the listicle work. (15) |
| 7 | Tue Oct 6 | circle.so/mcp updated with the connect steps for every assistant, and agents.json published beside llms.txt. (24) The "entire online business" post rewritten so we're in the recommended set. (15) | |
| 8 | Wed Oct 7 | Product updates, first step: every release on its own circle.so page with the full note, dated markup, llms.txt and sitemap entries, with web eng. (10) Best membership platforms and best community platforms posts: our entry rewritten. (15) | Web eng: every release on its own circle.so page. (10) |
| 9 | Thu Oct 8 | Retire, redirect and cleanup ticket shipped: /academy, /demo-video, the internal note, the stubs, the event-hub noindex, the eight 301s. (3, 4) Best creator platforms, event and membership-site posts: our entry rewritten. (15) | Web eng: the redirects, and noindex on the event-hub transcripts. (3, 4) |
| 10 | Fri Oct 9 | Price and fee table on the Skool, Mighty and Kajabi compare pages. (14) The Kajabi-alternatives verdict rewritten. Blog sweep batch one: ten posts. (15, 16) | |
| Week 3: term pages start; feature pages get 2026 on them. Claims wait on the open product facts. | |||
| 11 | Mon Oct 12 | Payments page: upsells, installments, Apple and Google Pay, fee math. (18) Mastermind page rewritten on its existing URL, as a platform page. (17) | |
| 12 | Tue Oct 13 | Events page: public RSVP, webinars, paid tickets, guest-to-contact; "webinar" in the title. (18) Challenge page, new. (17) | |
| 13 | Wed Oct 14 | Workflows page: sale triggers, Circle AI writes the workflow. (18) Accelerator page, new. (17) | |
| 14 | Thu Oct 15 | The /platform template fixed: FAQ schema, closing band, migration FAQ, the 2026 block. New-in-2026 blocks on seven feature pages. (13, 10) Signature program page, new. (17) | Web eng: the new-in-2026 blocks on seven feature pages. (10, 13) |
| 15 | Fri Oct 16 | The bios on every social and directory profile rewritten to the one-line description, if it exists by now. Otherwise a catch-up day. (8) Paid club page, new. (17) | |
| Week 4: the rest of the term pages; the two template pages land. | |||
| 16 | Mon Oct 19 | Kajabi compare page rewritten as the digital business page. (19) Discord alternative for creators page, new. (17) | Web eng: the help-center HTML mirror starts. (25) |
| 17 | Tue Oct 20 | Email hub, analytics, customization and gamification FAQs fixed; the 30-day challenge named. (18) Group coaching page rewritten. (17) | |
| 18 | Wed Oct 21 | Website builder page live, rebuilt with Rudy. (22) Cohort page rewritten. (17) | Rudy: design on the website builder page. (22) |
| 19 | Thu Oct 22 | Coaching page live. (20) Membership site for podcasters page, new. (17) | |
| 20 | Fri Oct 23 | Sales pages page live. (21) Blog sweep batch two: ten posts. (16) | |
| Week 5: the comparison layer and the selling pages. | |||
| 21 | Mon Oct 26 | Upsells page live. (21) The Skool comparison page rewritten. (19) | |
| 22 | Tue Oct 27 | Funnels page live: webinar, RSVP, sequence, checkout, upsell. (21) The Mighty comparison page rewritten. (19) | |
| 23 | Wed Oct 28 | Product updates, second step: per-release URLs and capability-style entries. (10) The Kajabi comparison page rewritten, and retitled off "Community Platform Comparison." (19) | Web eng: per-release entries and links. (10) |
| 24 | Thu Oct 29 | "What Circle sells that X can't" block and FAQ schema on every compare page. (19) /compare/circle-alternatives, new. (19) | |
| 25 | Fri Oct 30 | First-party reviews page live. (19) Blog sweep batch three: ten posts. (16) | Product: sizing for public community posts readable by crawlers. (26) |
| Week 6: the business-type family, and the shared pieces written once. | |||
| 26 | Mon Nov 2 | Creator business-type page. (23) High-ticket program page, new. (17) | |
| 27 | Tue Nov 3 | Education business-type page. (23) "A webinar that sells a course" page, new. (17) | |
| 28 | Wed Nov 4 | Membership business-type page. (23) Blog sweep batch four: ten posts. The tail of the 56 carries into November. (16) | |
| 29 | Thu Nov 5 | Community business-type page. (23) Every term page and story cross-linked; the six new stories linked from their term pages. (11, 17) | |
| 30 | Fri Nov 6 | Nav, trust line, title tags and the /platform hub, all written from the one-line descriptions you are already working on. (23) Month-one scoreboard: the thirty-prompt check re-run against the Sep 13 baseline, next to the AI-referral numbers we already track. | Everyone: month-one scoreboard. December: the localized pages. The Wikipedia case starts whenever PR can pick it up. (27, 28) |
A few ideas for what to watch. I'd rather the three of you decide what the right reporting is.
I don't want to prescribe the reporting here. You'll know better than I do what's worth tracking once you're in the pages, so take these as a starting list and tell me what you'd keep.
Under the hood we're better than I expected in two places and broken in four. The fixes are mostly settings, not projects.
Quinn ran one more pass on Sep 23, this time on the plumbing rather than the words: how each AI crawler is treated, what our sitemap and machine files say, how heavy the pages are, what structured data we ship, and whether the records that tell a model "which Circle" exist. Here's how we stack up against the companies that do this well (Vercel, Stripe, Whop, Neon), where the pattern is: every page readable as plain markdown, a curated llms.txt, an MCP server that's listed, no bot challenges, light HTML, structured data with prices, and a clean entity record.
I know this is a lot of pages. But the first bucket is about a day of work, the first three together are a couple of weeks, it's all facts, and it fixes the sentences that are teaching every assistant on the internet the exact positioning we're trying to move away from. I'd love to see those gone before the new homepage goes up.
Mostly I want to know whether this feels right to the three of you, and what you'd change. Happy to go through any of it live. And if you think I've got the order wrong, or the whole shape of it, tell me. I'd rather you own the sequencing than inherit mine.
Andy
The appendix, with every table, the five changes and eighteen fixes written up in full, and every page's verdict, is at the appendix page.