The Brand Brain
It knows each brand before it speaks for it.
Every brand has a Brain: who it is, who it serves, what it offers, how it sounds, and the claims it never makes. AKETRA plans from it, and the Guard checks against it.
- Voice
- Warm, precise, never pushy
- Audience
- Travellers and residents who book direct
- Pillars
- The terraceLocal foodEvents
- Never says
- “30% off”“Best in the city”
Sample brands and data
How it works
- 1
Import it from the website
AKETRA reads up to 20 pages of the brand’s website: the page you give it, then the pages of the same site its links and sitemap lead to, as the site’s robots.txt allows, with their structured data and the site’s llms.txt when there is one. You can also have it read one page only. From each page it proposes up to twelve items, up to 40 in all: identity, positioning, audience, products, offers that end on a given day, value propositions, voice, content pillars and knowledge. Each one quotes the passage it comes from, as written on that page or in the llms.txt, and one whose quote AKETRA can’t find there is dropped.
- 2
Or give it your documents
Give AKETRA a PDF, Word, text or Markdown document of up to 100 pages and 10 MB, such as your brand guidelines or a product sheet. It keeps the document’s text, not the file, reads it in the background and proposes up to 40 items, each quoting the document. A scanned document has no text it can read.
- 3
Or let it learn from your past posts
Ask AKETRA to read the brand’s latest 100 posts your team published through AKETRA in the last 12 months, leaving out the ones AKETRA wrote. It proposes the content pillars they keep coming back to and how they sound, each quoting a post, and no facts: prices and offers in old posts go stale. Posts published outside AKETRA aren’t read.
- 4
Or build it with your AI app
Your AI app reads what AKETRA can’t, such as your codebase. Over AKETRA’s MCP server, its build_brain prompt proposes what it finds; in any other AI app, you copy AKETRA’s prompt and paste the answer back. A fact the Brain already has isn’t proposed again.
- 5
Approve what’s right
Proposed items do nothing until an admin accepts them, one at a time or several together; rules are accepted one by one. Your team can also add and edit items directly.
- 6
Keep it current
Anyone on the team can dispute a fact: AKETRA stops using it until an admin keeps or retires it. Updating a fact keeps the earlier version, what members write is marked unverified until an admin confirms it, and an offer stops being used after the day it ends. When a new import from the website changes a number in a product, an offer or a fact in use, such as a price, AKETRA stops using it, proposes the website’s version and asks which is right.
- 7
Set the rules
Prohibited claims, guardrails and approved claims come from your team, never from the website. Your AI app can suggest them, and only admins and the brand’s approvers accept or change them.
- 8
Plan from it
Every plan reads the brand’s rules in full and as much of the rest of its active Brain as fits a set size. Each post it proposes shows the facts it relies on, or that it cites none, and what it assumes beyond them. Without the brand’s positioning, audience and voice in use, or while one of them is disputed, AKETRA doesn’t plan: it asks for what’s missing. The Guard holds any post that contains a prohibited claim.
What stays with you
- Nothing proposed is used until an admin accepts it
- Only admins and the brand’s approvers set prohibited claims, guardrails and approved claims
- A disputed fact isn’t used until an admin decides
- AKETRA asks for what the Brain lacks before it plans
- When your team rejects a post, the reason is proposed as a learned preference
- Every change to the Brain is on record
Questions
- What does AKETRA read from our website?
- The public page you give it, over HTTPS, and up to 19 more public pages of the same site that its links and sitemap lead to, never one its robots.txt disallows, and nothing past the page you gave when the robots.txt can’t be reached. Once you’ve set the brand’s languages, it keeps to them, as the site marks its pages: a page in another language is left out, and so is a translation of a page the site also has in the brand’s default language. It reads up to 10 MB of these pages in all, their text and structured data, and the site’s llms.txt if it has one, without running anything and never behind a login. It proposes what the site says and is told never to invent anything. Each item quotes the passage it comes from, so you can check it before accepting it.
- How is a prohibited claim checked?
- Each prohibited claim is a phrase, with the other ways your team says it can be written. The Guard holds any post AKETRA proposes that contains one of them, whatever its letter case, accents or Arabic letter forms. A rewording nobody listed isn’t caught yet.
- Is our Brain used to train AI models?
- No. AKETRA doesn’t train AI models on your content. It sends the brand’s active Brain as context with each plan and each website import, up to a set size, with the rules always in full.
More of AKETRA
Start with the Brain.
For each founding brand, we set up the Brain and the policy with you.