7. Footnote

This past fortnight I’ve spent a bit of time building out some of the data review and discovery processes that accompany PIPI, which is acting as a bit of a training ground for the PhD.

The big one is a new tool that I have to automate some of my previous coverage assessment processes. Using the DataforSEO API, the tool queries Google for coverage of assigned local government areas using operators to restrict results to the site in question; the council name and the three largest towns; and a fixed year fixed at the end of the previous month and assessed quarterly. I get back 20 results per quarter, including the result rank, the headline, a snippet, the link, and an initial assessment of whether it clears the geographic eligibility threshold.

When reviewing the Southern Cross Media regional newspaper Countryman’s coverage of Cunderdin LGA, for example, the relevant operators on the search query for a September/October/November quarter are site:countryman.com.au (“Cunderdin” OR “Meckering” OR “Warding East”) after:2025-09-01 before:2025-12-01.

That results in a summary table which provides the relevant quarter, the query string and ways for me to test the results manually against what was returned through DataforSEO, the number of results that were returned and which seem to clear the threshold, both by quarter and by month, and an initial assessment of whether they appear to be ‘strong’, ‘moderate’ or ‘weak’ results, or whether they’ve been excluded or are otherwise unclear.

The full list of returned results for each query are then provided, with the position on page, the date, headline, snippet, link, and the initial classification and reason. These latter two are really blunt: it’s really just asking whether one of the queried place names appears in the headline (strong) or the snippet (moderate). Weak is not assigned automatically. ‘Excluded’ is currently only automatically assigned to results that contain the words ‘funeral’ or ‘obituary’, because I was getting large numbers of results from page two that were death notices rather than journalism.

This process is already useful for me because at a click and for a few cents I can queue a quick assessment of any outlet in the PIPI data. It still requires a manual review of the results but greatly simplifies and speeds up that process.

The more exciting thing for me (and the one that I’m still tuning) is the discovery element.

When an LGA clears the coverage threshold, the tool then automatically submits a new query for every neighbouring LGA, whether or not it is currently assessed in PIPI as providing coverage of that place. It takes the same approach as above: search for the name of the council and the three largest towns or suburbs and assess a year divided into three-month quarters. If another LGA clears the eligibility thresholds then suggest it for inclusion and move to its neighbouring LGAs as well. If it fails, that branch of assessment ends.

This helps enormously with solving a longstanding problem. It is easy to review LGAs that are currently listed as covered, but it is hard to identify gaps in the record short of the time-consuming work of constantly re-testing coverage of LGAs that have previously found to be uncovered.

This approach unfortunately hasn’t been usable on sites that share a domain presence, like News Corp’s network of local titles, the majority of which are sub-sections of the metropolitan paper’s website.

It also created a new problem at highly networked papers like Southern Cross Media or Australian Community Media. Extensively internally-syndicated material across these companies means that the tool can end up pursuing coverage branches across half the country: I had the Western Advocate, ACM’s Bathurst paper, reach Victoria just by following the syndicated content chain from the Central Western Daily (Orange), Daily Advertiser (Wagga Wagga), and Border Mail (Albury). I cancelled that search rather than allow it to potentially go further.

That particular issue is what led to the ‘Jump’ column above, which allows me to set a certain number of adjacent LGAs from the initial set to test – don’t go more than one LGA beyond Bathurst.

That’s a crude fix that doesn’t allow for the very likely possibility that there are LGAs adjacent to LGAs adjacent to Bathurst that the paper really does cover. It’s also a short term approach: if the initial one-jump past Bathurst to Mid-Western Council clears the latter as a coverage area due to syndication from the Mudgee Guardian, then the next time I run the tool it will assess one jump past both Bathurst and Mid-Western to Upper Hunter, which may clear due to syndication from the Scone Advocate, and so on, and so on.

One potential longer-term solve for this might be to implement some of the theory that Kristy Hess and Alison McAdam have done around the effective local coverage radius of a local newspaper, and set a range of – say – 50km from a central point and assess LGAs that fall inside that circle. For Bathurst, that would capture Lithgow, Blayney, Oberon, Orange and Cabonne LGAs.

That might work in regional areas while not being suitable for metropolitan ones, and it introduces a new problem: how is that ‘central point’ determined? Sometimes it will be obvious: the Western Advocate are a Bathurst-based newspaper with a street address; sometimes it will be inferrable – The Irrigator gave up its office but is historically Leeton-based paper; but for a newer, region-wide, digital outlet like the New England Times? That’s less clear.