Case Studies How We Work Blog About Book a call
AI Services · MarTech · US NDA

Ascend tech: AI SEO recommendations inside CMS

Ascend.tech embeds in CMS to suggest SEO improvements and alt tags. Link suggestions hallucinated URLs; titles were keyword-stuffed and too long. Prompt engineering fixed both without changing core architecture.

Discuss a similar project All case studies

Service

AI Services

Industry

MarTech · US

Team

Backend · AI engineer

Technologies

Golang · PostgreSQL · REST API · LLM · Docker

Client context

Ascend.tech sells an SEO assistant inside client CMS instances. AI link and title features drove adoption but quality issues forced manual QA on every suggestion.

Challenge

Model invented URLs outside allowed list; anchors did not match destinations; titles exceeded limits and read unnaturally.

What we did

  • Link prompt: whitelist-only URLs, context-aware anchors, relevance scoring rules.
  • Title prompt: intent-aware keywords, anti-stuffing, hard character limits.
  • Regression checks on previously failing scenarios.
  • Go + PostgreSQL service layer unchanged; prompt layer upgraded.

Process

  1. Failure taxonomy from production samples.
  2. Prompt v2 draft and A/B on edge cases.
  3. Client-side validation of suggestion acceptance.
  4. Document prompt rules for future model swaps.

Result and impact

Link suggestions stay within catalog; titles usable without manual rewrite; lower QA burden on SEO workflows.

0 hallucinated URLs in production (post-fix)

This is an NDA-protected engagement. Client name and identifying details are withheld; industry, region, challenge, solution and outcome are shared in an approved form.

Suggestions we can actually publish — not edit into shape.SEO product owner (quote representative, under NDA)