Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.onsomble.ai/llms.txt

Use this file to discover all available pages before exploring further.

A scan produces numbers. Acting on recommendations turns those numbers into changes the business can actually make. This page covers how Onsomble’s recommendations are generated, how to prioritise them, and how to measure whether your changes are working.

​
How recommendations are generated

At the end of every scan, Onsomble analyses what the models said — and didn’t say — and produces a set of prioritised recommendations for improving how the business shows up. Recommendations are grounded in the actual responses. When a recommendation says “publish a dedicated comparison page for enterprise buyers,” it’s because the scan saw competitors winning on enterprise-specific prompts and the business’s existing content didn’t address that audience directly. Each recommendation includes:
  • What to do — the specific action being suggested
  • Why — the scan observations that prompted the recommendation
  • Expected impact — which metric or prompt set this is likely to move
  • Effort — a rough indicator of how much work is involved

​
Recommendation types

Most recommendations fall into one of a few categories.
A topic, audience segment, or question where the business doesn’t have content that AI models can draw on.Typical fix: publish a dedicated page, article, or section that directly addresses the gap. A well-structured 600-word page is often enough to shift citation on the prompts it targets.
Something the models are currently saying about the business is out of date or incorrect — pricing, service area, product lineup, team composition.Typical fix: correct the underlying source. If the incorrect information is on the business’s own website, update it. If it’s coming from a third-party source — a directory listing, a press release, a partner site — get the source updated or superseded by more recent, authoritative content.
The business is described in a way that doesn’t match how it positions itself — a specialist described as a generalist, or a premium provider described as mid-market.Typical fix: strengthen the positioning signal across the website. Explicit statements of specialisation, customer fit, and differentiation tend to carry through to AI descriptions when they’re consistent across multiple pages.
The information is there, but structured in a way that makes it hard for AI models to extract. Key answers buried in long paragraphs, important facts locked in images, or critical pages that aren’t well-linked from the rest of the site.Typical fix: restructure key pages — clearer headings, explicit Q&A formats, front-loaded summaries, improved internal linking.
Competitors are benefiting from reviews, case studies, press coverage, or directory listings that the business doesn’t have.Typical fix: the slowest but often highest-impact category. Investing in reviews, publishing customer stories, getting listed in trusted directories, and earning relevant press coverage all feed into the picture AI models build of a business.

​
Prioritising the list

Not all recommendations are equal. A practical order-of-operations:
  1. Fix inaccuracies first. Negative sentiment and factual errors are actively harming the business. They’re also usually the quickest to correct.
  2. Close the highest-leverage content gaps. Look for recommendations that would improve the business across multiple prompts at once. A well-structured service page often moves many prompts simultaneously.
  3. Work on positioning. Positioning improvements take a few pages of work but compound — they influence how the business is described across every future scan.
  4. Invest in structural improvements. These pay off over weeks as AI models re-index the improved content.
  5. Build long-term third-party signal. Treat this as a parallel programme rather than a quick fix. Start it alongside the other work.
If you’re doing this for a client, the first three categories are the most demonstrable in a follow-up review meeting. Structural and third-party improvements are longer-horizon and harder to show impact on in a single week or month.

​
Tracking impact

After making changes, let a scan run and then compare. Onsomble’s scan comparison views show how each metric — citation benchmarks, share of voice, sentiment — has moved since the previous scan. A few realistic expectations:
  • Content changes usually show up within one to three scans. The speed depends on how frequently the underlying models are updated with new web content.
  • Accuracy fixes can take longer if the incorrect information is cached in multiple sources. Correcting one source isn’t always enough — models draw from the web in aggregate.
  • Not every change moves every model equally. ChatGPT, Claude, Gemini, and Perplexity source differently. Expect uneven movement.
  • Negative movement after a model update isn’t necessarily your fault. Providers retrain and re-source regularly. A week of weaker scores that coincides with a model update usually reflects that rather than anything you did.

​
What’s next

The next chapter is Workflows — once the business is being described well, the next lever is making the business interactive through those same AI assistants.

Workflows overview

Turn knowledge into queryable content and processes like “get a quote” into structured workflows AI assistants can guide customers through.