Odd rankings, fewer homepage clicks and what ChatGPT counts as search

A hotel, a messaging app and a business loan have one thing in common in this week’s search data: the word “line”. The example came from a rank tracker. Further down, we look at what a ChatGPT reporting tool could show and where press coverage may fit into AI visibility reporting.

When a rank tracker finds a hotel in the banking results

In a study of 10,000 US desktop keywords, rank-tracking provider SE Ranking collected results for “business bank line of credit” that included a hotel and a messaging app. Both had a connection to the word “line”. Neither answered the banking question. Most of the blocks of unrelated results in the study started at position nine or lower. The researchers examined results collected by their tracking system, so the study leaves open what people saw when searching Google directly. [1]

That matters when a ranking report makes it look as though two of your pages are competing for the same search. You could spend time rewriting the wrong page if you act on the tracker alone.

I’ve had to check a similar mismatch while investigating whether two pages were competing for the same search. A tracker showed pages ranking for terms I couldn’t corroborate in Search Console or in a manual search. For my own checks, I start by filtering for the query in Search Console to see which pages recorded impressions and clicks. Check whether the apparent switch persists and use a live search in the relevant market as a second check.

What falling homepage clicks tell you about your brand

SEO manager Jenny Lee Lynch analysed 110 sites for Search Engine Land. On 81 of them, the homepage received fewer Google clicks between July and September 2026 than in the same period a year earlier. That finding raises a question for anyone hoping to see the effects of AI discovery in their homepage figures. If someone first hears about you from an assistant and later searches for you on Google, where would that interest show up? [2]

Similarweb’s research, which we covered in an earlier edition on AI reporting, helps explain why I’d ask. In its sample of US desktop users researching finance, travel and beauty, people were 2.5 times as likely to visit a recommended brand as a competing brand within seven days. That measures subsequent visits. It leaves open whether people searched for the brand on Google, landed on the homepage or went on to buy. The homepage study, meanwhile, measures clicks to one page per site and can’t establish why they fell. [8]

A referral report captures visits recorded from the assistant. If someone encounters the brand there and later goes looking for it elsewhere, that earlier encounter may be invisible in the report.

Where people get the answer

Someone looking up your opening hours might find the answer directly in Google and leave. Someone comparing your prices could land on a page deeper in your site, find a review on another domain or ask an assistant to compare the options. Those are plausible routes. The homepage figures alone can’t tell you which one people took.

I’d separate searches for the brand alone, searches for the brand with a qualifier and searches for what you sell that don’t name you. Use Search Console to see where each group lands on your site, then check the results for specific questions to see which outside sources are answering them. If visitors are landing on a service or comparison page, check whether it gives them a clear route to enquire and whether they take it. That’s more useful than trying to get them back to the homepage.

What counts as a search in ChatGPT?

OpenAI’s Adam Fry told Big Technology that ChatGPT was receiving roughly a fifth as many “search-like queries” as Google receives queries. The interview leaves the counting method open. [3]

A person can ask for suppliers, give a budget and then request a shortlist in the same chat. The comparison tells us little about how many people are looking to buy or how many would otherwise have used Google.

For a commercially important topic, I’d test a real question from a sales enquiry and a plausible follow-up. See which businesses make the shortlist once the buyer adds a constraint, such as a budget or a location, and what information the assistant uses to justify including them.

What would we want from a ChatGPT reporting tool?

Search Engine Land reported that OpenAI was testing business reporting tools for ChatGPT. [4]

The scope of that reporting matters because a brand can feature in an answer without its website being the destination. If an answer mentions your brand and links to a review, the publisher gets the link and you get the exposure. If somebody clicks through to that review, it still won’t appear as a visit in your website analytics. A report about links to your domain would give you a different picture from a report about where your brand appears.

If you test both signed-in and signed-out sessions, keep a record of which conditions you used and compare like with like. Repeating a selected question shows how often your business appears across those runs. If the platform eventually tells us what people actually ask, I’d use that to revisit the panel. Read Go Up’s guide to interpreting AI reporting.

Even a report that shows where your brand appeared would only capture part of how someone found you. A visit from Google can come after the AI research that put your business on a buyer’s shortlist.

How I’d report on a local launch

Imagine you earn press coverage for a service you’re launching in a new area. I’d put the coverage dates and links alongside the brand and service queries relevant to that area, organic visits and enquiries for the location page, and the sources that appear in AI answers to local questions. That keeps the comparison focused on the service you actually promoted.

If an assistant cites an article about the launch, you can see which source it used and what it says. If relevant branded searches rise around the same time, the timing is worth investigating. The launch itself, changes to the website and other activity could also have played a part. One practical use for the report is deciding where to act first. If answers repeatedly cite a directory with outdated information, correct the listing before deciding whether you need more coverage.

Decide what the report should help you change

Before adding a metric to a report, decide what you’d do if it moved. If you want more enquiries from a service page, check where visitors stop before they reach the form, then measure whether more of them enquire after the fix. If you want to improve how often your business appears in AI answers, start with the customer questions that matter and check which sources keep getting cited.

Talk to Go Up

If you’re deciding how to respond to a change in search or AI visibility, talk to Go Up. We can help you work out which figures explain the change and where to put the work.

Sources

1. SE Ranking, research on unrelated Google results, 3 November 2026.

2. Search Engine Land, homepage organic clicks analysis, 4 November 2026.

3. Big Technology, interview with OpenAI’s Adam Fry, 3 November 2026.

4. Search Engine Land, reported tests of business tools for ChatGPT, 6 November 2026.

5. Search Engine Roundtable, Google AI Mode continuation prompt, 6 November 2026.

6. Google Chrome, Lighthouse 13.6.0 release notes, 27 October 2026.

7. Google for Developers, PageSpeed Insights lab and field data.

8. Similarweb, study of visits following AI recommendations.