How the Moss method works, step by step

Moss, a coworking space in Venice Beach, Los Angeles, went from not being cited by ChatGPT to being cited, and from about fifth to first on Google for "coworking Venice Beach," after its general manager rebuilt the website herself with a method taught by Eric Grows Companies. The method is a loop: see the site the way Google and AI tools read it, fix what they cannot read, and check again. This post sets out each step and shows what changed on Moss's homepage, measured from a public copy of the page saved before the rebuild.

Published: October 22, 2026. Written by Eric Levine, founder of Cobalt Bloom.

Moss is not a Cobalt Bloom client, and Cobalt Bloom did not build Moss's website. Elyse, Moss's general manager, did the work after taking Build Your Website with Claude Code, an Eric Grows Companies course, in July 2026. Eric Levine runs Eric Grows Companies and also runs Cobalt Bloom, which uses the same method. The full case study is at https://ericgrows.com/case-studies/moss-venice.

What was wrong with Moss's website before the rebuild?

Moss had a strong business and a website that did not show it. Moss had more than 150 members and a 4.9-star Google rating, and people found it through word of mouth, Google and ChatGPT. When AI tools described Moss, they pieced the answer together from thin Moss pages and sites such as Yelp. The homepage title was "moss | home for your ideas," with no mention of coworking or Venice. Six of the site's 16 pages were duplicates left live. The phone number and hours appeared only inside the chatbot's code. The membership page showed two different price tables and still mentioned a launch party from 2023. From the front, the site looked finished.

Step 1: How do you see a website the way AI tools see it?

Run an audit that reads each page the way a crawler does, then scores what it finds. An AI crawler is the program an AI company sends to read web pages. Some AI crawlers do not wait for a page to finish loading, so anything that only shows up once the page has loaded can be invisible to them.

Elyse ran the audit on July 14, 2026, in the third class of the course. The audit came as a Claude skill, a saved set of instructions she could run again whenever she liked. It checked the page titles and descriptions, the text on each page, which pages existed, whether AI crawlers were let in, and the structured data. Moss scored 37 out of 100, with the verdict "Poor, but cheaply fixable." The audit sorted its findings from critical down to minor, so Elyse knew what to fix first.

Step 2: Does the page say what the business does and where, in plain text?

A business's homepage should say who it is, what it does and where it is, in words anyone can read on the page, and the page title should say the same. The page title is the name that shows on the browser tab and in Google's results. The phone number and hours belong in the page text too, not inside a chat window or a pop-up.

Cobalt Bloom compared Moss's homepage before and after the rebuild, using public copies only. The "before" is the Internet Archive's copy from July 11, 2026 (https://web.archive.org/web/20260711152239/https://mossvenice.com/), three days before the audit. The "after" is the live page on October 22, 2026. Both were read the way a crawler that does not run scripts would read them.

The old title said neither "coworking" nor "Venice"; the new title says both. That matches Cobalt Bloom's study of 100 local searches (https://getcobaltbloom.com/guides/why-isnt-my-website-showing-up-on-google#what-gets-your-website-onto-the-first-page), where most websites on Google's first page named their service in the home page title. The phone number was in the old page's code, but only as data for the chatbot, not in the text a crawler treats as the page.

Step 3: Does every fact appear only once, and agree everywhere?

Each fact about a business should have one version: one set of prices, one list of hours, one page per topic, and nothing out of date. When the facts disagree, an AI tool has to guess which one is right, or fill the gap from somewhere else.

Moss had six duplicate pages competing with the real ones, two price tables that did not match, and a launch party from 2023 still on the membership page. The rebuild cut the site down to eight pages, none of them duplicates, with one pricing table. This is the step most owners skip, because nothing looks broken. In Cobalt Bloom's check of 95 small-business websites (https://getcobaltbloom.com/guides/what-ai-tools-find-on-small-business-websites), more than half (56%) did not label what kind of business they are in their structured data. Only 15% turned away any AI crawler.

Step 4: Is the business labeled for machines?

Add structured data: a short block of code, invisible to visitors, that states the business's name, address, phone number, hours and type in a format search engines and AI tools can read without guessing. For a local business, the type is LocalBusiness or one of its more specific kinds. Moss's July 2026 homepage had no structured data. Moss's homepage in October 2026 has one block naming Moss as an organization and a local business, with its address, phone number and opening hours. For a business rebuilding too, Cobalt Bloom's redesign guide covers keeping Google rankings through the move, at https://getcobaltbloom.com/guides/how-to-redesign-a-website-without-losing-seo.

Step 5: What happens after the first round of fixes?

Run the audit again. Each run shows the next set of problems, including ones the fixes themselves created. Elyse used the same loop for every change: she ran the audit, gave the findings to Claude, made the changes, then ran it again. Rebuilding a site can break things that worked before: pages move to new addresses while Google still lists the old ones, and visitors land on a dead page. A re-run catches it, which is why the loop never really ends.

Moss's audit score went from 37 in July 2026 to 71 on September 9, 2026, on the same scoring rules. Eric Levine's part, in his words from the case study: "I didn't touch a thing. I taught the method and answered questions as they came up."

What results did Moss get?

The Moss case study reports that ChatGPT began citing Moss, Moss rose to first on Google for two searches, and weekly tour bookings doubled with no ads. The audit and Google positions were measured between July and September 2026. The tour figure is Moss's own, covering spring to September 2026.

Cobalt Bloom checked again on October 21, 2026: Moss was named in 80% of AI answers across three searches. The details are in the guide to getting recommended by ChatGPT, at https://getcobaltbloom.com/guides/how-to-get-your-business-recommended-by-chatgpt.

What does the Moss result show?

Moss is one business, so it shows what the method can do rather than proving it works the same everywhere. Search results and AI answers move from week to week, and Moss already had strong reviews, a real community and people looking for it. The website was the piece that changed. Cobalt Bloom's test of 100 "best [business] in [neighborhood]" questions points the same way: businesses with a website on their Google listing were named 1.6 to 2 times as often as those without one. Cobalt Bloom's guide to Google's AI Overviews found the same for Google's AI: in its October 2026 Google searches that showed an AI Overview (48 searches), 94% of the businesses AI Overviews named had a website on their Google listing, against 74% of those passed over (https://getcobaltbloom.com/guides/how-to-get-your-business-into-google-ai-overviews). The audit score is Cobalt Bloom's own scoring rubric, not a number Google or any AI tool uses; it is useful for tracking one site over time.

What did Moss's rebuild change behind the scenes?

Most of the work was behind the scenes. Moss's rebuild added structured data to six of its eight pages, removed the duplicate pages and cleaned up the sitemap, the file that lists a site's pages for search engines. It added an llms.txt file that describes Moss for AI tools, and checked that every major AI crawler was let in. Elyse did the rebuild herself, with Claude, over a summer.

Most owners do not have a summer to spend on their website. Cobalt Bloom takes that work on for them, using the same method and keeping it up every month. Cobalt Bloom has not launched yet; its waitlist is open at https://getcobaltbloom.com/#waitlist.

How is getcobaltbloom.com using the same steps?

getcobaltbloom.com is running the same loop on itself, from a harder start. The domain was registered on October 10, 2026, it has no reviews, and it shares its name with a mineral. On October 21, 2026, no AI tool cited it for any of 15 owner questions. The first post, at https://getcobaltbloom.com/blog/starting-from-zero, has that starting point and the 90-day plan. Each month Cobalt Bloom re-runs the same check and reports the numbers.