ChatGPT stopped quoting last year’s pricing

The problem

Tallyhaus had changed how it charged. In early 2025, the expense management company replaced per-user plans with one flat team price. Its current pricing page reflected the change, but the rest of the web had not caught up.

A year later, 62% of the AI answers we checked still quoted the retired per-user model. Buyers were arriving at sales calls with expectations the team could not meet. The problem was more specific than low awareness: people had heard of Tallyhaus, but some had learned the wrong commercial facts.

Publishing another announcement would not resolve all the contradictory versions. We needed to identify where the old numbers were still visible and give both people and search systems a clearer current reference.

Defining the correction

We treated the pricing model as a set of facts to verify, rather than a single number to replace. A useful answer had to distinguish the billing unit, the plan, what the team price included and any limits that changed the cost.

The measurement stayed focused on answers quoting old pricing. A response that simply omitted a price was logged separately; silence should not be mistaken for a correct explanation. The same pricing questions were checked throughout the ten-week engagement.

What we did

Weeks 1–2. We found 31 pages and listings still carrying old prices. The inventory included two review sites and an archived help article. For each source, we recorded the incorrect statement, the current replacement and the person who could update it.

This gave the work an order. Pages repeatedly cited in wrong answers went to the top of the queue. A stale page with no observed connection to the tracked questions could wait behind a source that was actively reinforcing the mistake.

Weeks 3–6. We rebuilt the pricing page around one plain table covering plan names, limits and prices. Supporting copy explained the team-based model directly. The help center used the same terms, so a buyer following a second link would not encounter another interpretation of the offer.

Weeks 7–10. We followed through with review sites and partner directories and had the archived article removed from search. Correction requests linked to the current reference and identified the exact outdated statement, making the update easier for the other publisher to check.

We continued testing while those requests were in progress. Completing an outreach email was a delivery milestone; seeing the source change and the answer correct itself were separate milestones.

What moved

Answers quoting old pricing fell from 62% to 3% in ten weeks. ChatGPT changed first, roughly four weeks after the new page went live. Perplexity followed as the review-site listings were updated.

The sequence was consistent with the sources we were correcting, but several changes overlapped. We would not assign the full improvement to a single table or listing. The useful outcome was a much more consistent explanation across the tracked pricing questions.

What we’d do differently

We would contact the review sites on day one. Their listings were among the most frequently quoted sources and the slowest to change because their publishing schedules were outside Tallyhaus’s control.

The broader lesson is to include external listings in a pricing launch. A new page can be correct while the old offer remains easier to find elsewhere. Build the correction list alongside the launch checklist and keep checking it after the announcement is over.