Site memory
The five layers of learned context Contexta injects into every AI prompt so suggestions sound like your brand and improve over time.
Site memory
Generic AI output is the reason most AI writing tools get abandoned. Contexta carries persistent, site-specific context into every prompt so suggestions sound like your site — and get closer to it the more you use the plugin.
Find it at Contexta → Settings → Memory.
Layer 1 — Site DNA
The foundation you set once. Four fields, each with presets or your own text:
- Tone — educational and friendly, professional and authoritative, warm, bold, luxury, playful, technical…
- Register — how formal the writing should be, offered in your site's own language
- Niche — what your site is actually about
- Market — who you sell to and where
Your brand name always follows Settings → General → Site Title, so it's never out of sync.
Your site tagline is a good starting point for the niche field — Contexta shows it while you edit.
Layer 2 — Product catalogue
For shops, Contexta syncs your products with their live prices and uses them when writing product CTAs.
This is what stops the AI inventing a price or recommending something you don't sell. Prices are read at generation time, so a CTA never quotes a stale figure. Click Sync Product Catalog after big catalogue changes.
Relevance matters here: for any given article, Contexta picks the products actually related to the topic rather than whatever is newest.
Layer 3 — Approved examples
Every time you save an AI-generated title or meta description, Contexta records what it suggested and what you kept.
Those approved examples are fed back into later prompts as concrete evidence of what "good" looks like on your site. This is the layer that makes week four noticeably better than week one — you are training it simply by working normally.
Layer 4 — Content cluster map
Contexta builds a map of your topical clusters — which articles belong together and how your content is organised.
New content is written with awareness of what already exists, which reduces accidental overlap and helps internal linking suggest genuinely related pages instead of superficially similar ones.
Click Build cluster map to refresh it after publishing a batch of new content.
Layer 5 — Rejection signals
The mirror of layer 3. When you heavily edit or discard a suggestion, Contexta notices the difference and records the pattern.
Phrasings you keep rejecting stop being suggested. You never have to explain why — the edit itself is the signal.
How it's used
All five layers are condensed into a context block that is added to every AI prompt: title generation, meta descriptions, intros, sections, FAQs, rewrites, CTAs. There's nothing to attach manually.
The block is size-limited so it never crowds out the actual task — adjustable through the
rm_memory_char_budget filter if you have an unusual setup.
Managing memory
Each layer can be cleared independently from the Memory panel. Useful when:
- You rebrand and the old tone no longer applies
- You imported content that taught it the wrong patterns
- You want to start the approved-examples layer fresh after changing direction
Clearing one layer leaves the others intact.
Privacy
Memory lives entirely in your WordPress database. Nothing is uploaded anywhere. The only data that reaches Anthropic is the context block attached to a generation you explicitly requested — there are no background calls.