Scored from your page.
An agent that fixes it.

Byline reads a page the way a crawler does and scores two things. Retrievable is a score out of 100 for whether an engine can reach the page, parse it and lift a clean passage out of it. Citable is a band for whether a lifted passage carries anything that points back to you, such as a named author or data only you publish.

Every finding shows the evidence it was scored from, down to the heading text as found or the line in your robots.txt that decided a crawler's access. When too little of a page can be read, Byline publishes no score rather than a guess. The same rubric scores drafts in the editor before you publish them.

Because citations move when nothing on the page changes. Otterly.AI reported that between July and September 2026 Reddit's share of ChatGPT citations fell 94% while its share of Perplexity citations rose 86%. In Semrush's 30-day test, Google AI Mode cited 48 of 81 unchanged pages on day six and 21 on day thirty.

A single visibility number averages those movements into something that happened to nobody. What stays still, and what you can change, is the page. So the score comes from deterministic code reading your HTML. A language model writes the explanation beside a finding and nothing else. It cannot set a score, a status or a band.

Sources: Otterly.AI, 1 October 2026 and Semrush, 10 December 2025.

Otterly.AI, an AI search monitoring vendor, also publishes a content readiness score. Its public API describes that score as 13 named dimensions, such as specificity and paragraph cohesion, each returned as a number from 0 to 100 with a free-text explanation.

The schema has no field for text quoted from the page, and it does not say how the dimensions combine into the overall score. Otterly also runs prompt monitoring across AI engines, which is a different product; this is about its readiness score only. The schema is public, under PublicApiContentCheckContentAnalysis in Otterly's API specification, checked on 11 October 2026.

Each Byline finding names its threshold and quotes what it found. 'Self-contained sections 0/8: no section pairs a heading with enough prose to stand alone (target: 8+ sections of 40+ words)' is a line you can check in a browser in a minute. Factors that cannot be checked are marked unable to assess and left out of the total, never scored as zero.

Every report records the engine version that produced it, currently 2026.12, so two runs can be checked for whether the rubric moved between them. A language model never decides a score, and the score is withheld when the page cannot support one.

In October 2026 we audited the homepages of eight SEO agencies in the UK and US. All eight could be read well enough to score, between 62 and 79 on engine 2026.10. Wherever the crawler check could be attributed, every AI crawler was served the page. On two sites an ordinary browser request was refused too, so every crawler row there was reported as unknown, not blocked.

The gaps were on the page. Seven of the eight had fewer than eight sections that stand alone when quoted. Four had no heading phrased as a question. One had its main Organization markup fail to parse because of a line break inside a text string.

Those runs also exposed errors in our own engine: a CAPTCHA script on a contact form read as a bot challenge, and a legal company name searched for instead of the trading name. We fixed both and re-ran all eight before using any finding.