How we work

A specialist you trust. AI that works.

You talk to a marketing specialist. They run an AI team that does the legwork. Nothing goes live until you say yes.

Working with us.

A free call

Your business, your customers, your budget. You leave with a plan whether or not you hire us.

Set up together

Your specialist connects your website, Google and ad accounts with you, on a call. You never handle a key.

The AI team works

Research, fixes, drafts and campaigns, day and night. Your specialist checks what reaches you.

You approve and watch

Say yes to what you like. Your dashboard shows every piece of work, and what it did.

Prefer to do some of it yourself? The tools your specialist uses are open to you in Thymelab, our do-it-yourself lab.

What the AI does to your website, all nine steps
Step 1

You give it an address

And, optionally, context that sharpens everything after it: what kind of business this is, where it operates, who it competes with, and the terms you want to win. None of it is required.

Step 2

It reads the site's own instructions first

robots.txt, the sitemaps named in it and the ones at the usual paths, and llms.txt. The AI crawler verdict is decided here, before a single page is fetched.

Step 3

It crawls

Breadth first from the homepage, with sitemap URLs entering at the same depth as navigation links so orphan pages surface rather than being starved. robots.txt is honoured. Redirect chains, status codes, response times and payload sizes are all recorded on the way through.

Step 4

The catalogue runs

Per page and across the site: duplication, cannibalisation, orphans, sitemap accuracy, entity signals. Each finding gets a priority from impact times confidence over effort, scaled by how many pages it touches, so the list is ordered by what is worth doing rather than by severity alone.

Step 5

The fixes get written

Titles from the page's own topic. Meta descriptions from its own opening. JSON-LD from what it contains. Sitemaps from the URLs that were reachable and indexable. Link plans from pages that already discuss the target. Alt text from filenames and page context, with a warning to read it.

Step 6

Strategy, not just repairs

A keyword model built with tf-idf over your own copy, weighted towards titles and headings. Gaps against your competitors, your locations and the terms you named. Briefs with outlines, answer blocks, FAQs, internal links, meta and schema.

Step 7

You approve

Batched by risk, each item showing what it changes, why, what it is expected to do and whether it can be undone. High and critical items cannot be bulk approved, deliberately.

Step 8

It ships, or it hands you the exact change

Through your CMS where one is connected. As a webhook payload, a patch or plain text where one is not. Either way the change is recorded against the agent that proposed it and the person who approved it.

Step 9

The next run reports the difference

What cleared, what appeared, what got worse, which pages changed, which way the scores moved. That comparison is the thing an agency retainer is supposed to produce and rarely does.

What gets connected.

Your specialist connects these with you. Each one lets the team do a bit more, and each can be switched off from your own account at any time.

Measurement (3)
Recommended

Google Search Console

The only first-party record of what people searched before they reached you.

  • Real query data instead of a keyword model derived from your own copy
  • Indexing status of your top pages, and whether Google agrees with your canonical
  • Click and impression movement attributed to each change

Without it: Keywords are inferred from the site's own content, and results cannot be attributed to the work.

Recommended

Google Analytics 4

Sessions, conversions and revenue, so a ranking change can be tied to money.

  • Conversions and revenue per landing page
  • Before and after reads on every change that shipped
  • Traffic segmented by the channel that actually brought it

Without it: Reporting shows crawl and ranking movement but cannot say what it earned.

Optional

Bing Webmaster Tools

Bing feeds Copilot and DuckDuckGo, and almost nobody checks it.

  • Bing query and indexing data
  • IndexNow submission so changes are picked up in minutes

Without it: Bing and Copilot visibility is unmeasured.

Local and maps (1)
Optional

Google Business Profile

The profile that decides whether you appear in the map pack.

  • Weekly posts written and queued for approval
  • Review replies drafted within hours, gated by rating
  • Hours, categories, services and attributes kept complete

Without it: Local findings are limited to what the website itself shows.

Where the site lives (4)
Optional

WordPress

Publish approved content and apply on-page fixes directly.

  • Approved drafts published as posts or pages
  • Title, meta and schema written back to the page
  • Internal links inserted into existing copy

Without it: Every approved change is handed to you as the exact text to paste, with where it goes.

Optional

Shopify

Product, collection and blog optimisation written back to the store.

  • Product titles, descriptions and metafields updated
  • Collection pages given real copy instead of a product grid
  • Product schema completed with the merchant listing fields

Without it: Changes are produced as text and CSV for you to import.

Optional

Webflow

CMS collection items and page settings updated in place.

  • Collection items created and updated
  • SEO fields written per page
  • Publishing to the live site on approval

Without it: Changes are produced as text to paste into the Designer.

Optional

Any other CMS or static site

Git, a webhook, or a person. All three are supported.

  • Changes exported as a pull request, a patch file or plain text
  • A webhook fired with the approved payload

Without it: Nothing. This is the fallback that means no site is unsupported.

Search data (1)
Optional

DataForSEO

SERP positions, keyword volumes and a backlink index, pay as you go.

  • Real search volumes and difficulty instead of a derived model
  • Daily rank tracking for the terms that matter
  • The referring domain set for you and every competitor

Without it: Keyword work runs on Search Console data and the site's own topic model. Link work is limited to prospects the crawl can see.

Link intelligence (1)
Optional

Ahrefs

Referring domains, anchors, and what competitors earned recently.

  • Full referring domain profile
  • Lost and gained links
  • Competitor link gaps

Without it: Link prospecting works from the site's own outbound links and missing entity records.

Outreach (1)
Optional

Your email, for outreach

Outreach is sent from your domain, through your provider. Never ours.

  • Personalised outreach sent under a per-domain daily cap
  • Replies tracked against each prospect
  • Follow-ups that stop the moment someone answers

Without it: Outreach emails are drafted and handed to you to send.

You choose how hands-on to be.

From “ask me everything” to autopilot. Some things always need you, whatever you pick.

The autonomy levels
LevelWhat it means
ObserveAudit and report. Nothing is proposed for approval.
ProposeEverything is queued for you, including one-line meta changes.
AssistReversible, self-verifying changes auto-approve after a delay. Anything in your voice waits.
OperateTechnical fixes and schema ship on their own. Content and outreach wait for you.
AutopilotEverything reversible ships. Site-wide and irreversible actions still need a person, always.

Content written in your voice and outreach sent from your domain wait for a person at every level, including autopilot. That is not a default you can change. See how data and credentials are handled.

Tell us about your business.

Thirty minutes with a specialist. You leave with a plan, whether or not you hire us.