The six dimensions below come from the AI readiness check Avid built for media owners. The research figures carry their own houses, dates and samples.
The authority score sits on the one-pager next to monthly uniques, and in a rate conversation it is doing real work. It says: this is a serious title, so the engines will treat it as one. The first half of that is true. The second half has now been measured, and it is not.
Surfer, Joshua Hardwick, 14 July 2026, took three months of its own AI citation tracking, roughly 5 million unique citation source URLs from 20,000 prompts, and looked for a relationship between a domain's authority score and whether an AI answer cited the page. Across the full sample it found effectively no correlation between a domain's authority score and whether AI cites it, Spearman -0.073 to +0.026. The market the sample covers is not stated. Read the analysis.
Spearman runs from minus one to plus one. A range that crosses zero by a few hundredths in either direction is not a weak relationship, it is the absence of one. It stays near zero on every engine in the sample, and it stays near zero when the largest domains are taken out of it, which is the cut where a size effect would show itself if there were one. Narrower cuts of the same data wander further from zero in both directions, so read that headline range as the finding for the sample as a whole rather than as a law.
It does not say your authority score is a bad number. It says the score is a number about something else. Put on a rate card as evidence that a title will be cited, it is evidence of nothing of the kind.
An authority score estimates how likely a domain is to rank in a list of links, mostly from who links to whom. While the click was the product, that was a sound proxy for commercial value.
68% of Google searches now end without a click to a website, up from 60% two years ago. That is SparkToro, using Similarweb clickstream data, June 2026, in the United States. Read the study.
So the score describes your standing in the part of the market that is contracting, and stays silent on the part that is growing. Keep it on the one-pager. Just stop asking it to carry the half of the conversation it cannot reach.
Our AI readiness check scores a title across six dimensions. Three things decide whether a page gets quoted at all: whether it can be read, whether it carries a claim worth quoting, and whether the same claim turns up elsewhere. Each is a property of the page, or of the campaign around it, rather than of the masthead above it.
Two dimensions sit here, readability and access. Both are preconditions rather than achievements, and both fail quietly. The full article text has to be in the HTML on first load rather than assembled by scripts afterwards. Every article needs one stable address. Article and Organization structured data, with author, date published and date modified, has to be there and has to validate. No engine publishes that markup as a condition of being cited, and we are not claiming it is one. The check asks for it because it is the cheapest way to state who wrote a page and when, in a form a machine cannot misread.
Then robots.txt has to admit the crawlers that fetch a page in order to answer a live question. Those are not the same agents that collect text to train a future model, and the check treats them as two decisions because they are set separately.
Take the training decision provider by provider, and read the provider's wording rather than the category. Google's own token is the one that catches people out. Google-Extended covers whether content Google crawls may be used to train future Gemini models, and Google states that it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search". The same page also puts grounding in Gemini Apps, and grounding with Google Search on Vertex AI, under that token (Google Search Central). So blocking it is not free in answers: it keeps your pages out of Gemini's training and out of the answers Gemini grounds, in one move. That page says nothing either way about AI Overviews, so do not read a Google-Extended decision as a decision about those.
A title that blocks the answering crawlers site-wide scores zero on access, and the check caps the band there however strong the other five dimensions are. Nothing else the title does makes up the difference.
Models quote what they can check. Mood and story move people. A specific claim moves an answer. In practice that means one provable thing in every commissioned piece: a number, a comparison, a named product, a date. The headline is the question a buyer would ask, and the first paragraph answers it, rather than a line written to earn a share on social. The byline comes from someone who knows the subject, with real credentials on an author page, and the date is shown and updated when the piece is refreshed.
None of that is an engineering project. It is a brief template, which makes it the cheapest item on this list to fix.
One article in one title is a single data point, and a model can weight it or ignore it. The same claim carried by several unaffiliated titles inside one window reads as the settled view of a category instead of one publisher's opinion. This is the dimension a title controls least on its own, and it is why a campaign is briefed to four publishers rather than one. It also changes what is worth asking in a rate conversation. Not how big are you, but can you carry a claim in your own voice, to your own standard, inside an agreed window. One brief going to several titles at once is the shape the work takes.
Take two pieces on the same product category, published the same week. The first, in a large general-interest title, says the category leader offers the best value on the market. The second, in a small specialist title nobody outside the category has heard of, says which six models were run, for how long, under what conditions, what was measured, and which one came out ahead on each measure, under a byline with the credentials to have done it.
Only one of those is quotable. A model asked what to buy can lift the second and stand behind it, because the claim arrives with the conditions that produced it. The first is a verdict with nothing underneath it. A model can read that page and still has to leave it, because there is nothing in it to hold up against a rival's.
The specialist title wins that exchange on a lower authority score, less traffic and a smaller team. It wins because it published its method. Be clear about what that last paragraph is: nobody has published a measure of how much more often a page carrying a stated method is cited than a page carrying only a verdict, so it is reasoning, not evidence. What the measured finding establishes is the narrower half of it. Size does not decide which pages get quoted, so whatever does decide it is on the page, and the method is the part of a page a model can stand behind.
Three things, in the order an advertiser will care about them.
All three are harder to produce than a score anyone can look up, and that is why they are worth more in the room. A number the advertiser can pull themselves in ten seconds is not an argument. A citation record on their category is. The same logic runs on the brand side of the table: an answer is assembled out of pages the brand does not own, which is what makes a publisher's own citation record the thing being bought.
The AI readiness check is eight questions and about three minutes, with no sign-up. It returns a band, a score on each of the six dimensions, and the next steps for whichever two are weakest. If you would rather talk it through, or see where your title fits in a campaign running across several publishers, talk to us.
10 September 2026 · 6 min read
Before any engine can quote your title, the page has to clear conditions that have nothing to do with the writing. The six that decide it, the two halves of a quotable page, and the crawler distinction most robots.txt files have never made.
10 September 2026 · 6 min read
OpenAI's own documentation says a sponsored placement sits below the answer and cannot change it. That makes the ad and the recommendation two different purchases, out of two different budgets.
9 September 2026 · 6 min read
The largest kind of source in an AI answer is often one nobody can buy. Here is the subtraction that shows what is actually left to commission in your category.
See how your brand appears in AI answers, and where trusted publisher content can strengthen its visibility.
