BERT Relevance: our tool uses AI to decode landing page experience in Google Ads

In Paid Search advertising, it’s essential that your customers’ search terms closely match your landing page content. Google assesses this alignment through the ‘landing page experience’ metric, which includes elements such as content relevance, user navigation ease, and internal linking. With inefficient landing pages, brands will suffer from low Quality Score and increased costs.

That’s why our tech team created BERT Relevance, a tool that uncovers mismatches between search terms and landing page copy. This allows brands to make strategic decisions on improving landing page copy, generating new content, or removing irrelevant search terms. Ultimately, this will improve paid search performance and increase ROI.

Content relevance is crucial. ‘Landing page experience’ significantly influences a keyword’s Quality Score, which in turn affects the cost per click (CPC). Google categorises your ‘landing page experience’ score as below average, average, or above average. Given the wide range of factors this metric encompasses, these broad categories offer limited insight, often leaving analysts uncertain about the necessary actions.

BERT Relevance demystifies this metric, with a particular emphasis on content relevance. Our goal in creating this tool was to uncover mismatches between search terms and landing page content, allowing us to make strategic decisions on improving landing page copy, generating new content, or removing irrelevant search terms.

Which insights are uncovered?

Our tool BERT Relevance gives insights that guide strategic decisions in campaign management, from refining ad content and keywords to optimising landing pages and segmenting campaigns based on specific themes or services. These insights enable marketers to improve:

Landing Page Optimisation: Enhance pages with low BERT Relevance scores to better align with search terms or redirect ads to more relevant pages.

Query-to-Page Relevance: Ensure alignment between the search term and the landing page. Discrepancies can lead to low BERT Relevance scores and a poorer user experience.

Search Query Refinement: Reassess the effectiveness of search terms in driving traffic at either search query or query cluster level, identifying poor performing search queries and search query themes.


The workflow



  1. Search query report from Google Ads: The process begins with the collection of search queries, performance data and the landing page URL.
  2. Keyword Clustering: We then cluster the search queries from Google Ads using OpenAI and GPT4. Clustering this data makes it easier to extract insight across a very large data set (10k search queries at a minimum).
  3. Content Scraping via Puppeteer: We scrape the content from each landing page URL using Puppeteer, a nodejs library for javascript enabled web crawling.
  4. Content Classification: We then use our BERT Relevance model, based off a machine learning model used for sentence/text embedding generation. This assesses the alignment between landing page content and search terms – the output of this is our ‘BERT score’ used to define the relevance of the search term vs. the landing page.


How our clients have benefited

  • Account Restructuring: When onboarding a new client that previously had a fragmented account structure, we used insights from a long term BERT Relevance report to determine the most relevant landing page to use for each ad group within the new account structure. The new account structure resulted in a 19% increase in impressions being classified as ‘above average’ for the Google landing page relevance score, with an average quality score increasing from 6.5 to 9.2.
  • Search Query Funnelling: For a retail client with a large product inventory, the report identified search queries which were triggering multiple ad groups/landing pages. The BERT Relevance report provided insights that informed us where to add negative keywords to ensure that the most relevant landing page is targeted for each search query. This resulted in both more relevant adverts, which recorded a 6% increase in CTR%, and an 8% improvement in conversion rate for associated keywords.
  • Blocking Poor Performing Traffic: We now use the report on a regular basis to identify poor performing search query ‘clusters.’  This groups search queries into similar themes. In isolation, poor performing search queries may go unnoticed. However, the clustering approach has proven to unearth poor performing one or two word themes, which we have used to add as negative phrase keywords to block irrelevant traffic.
  • Informing DX/CRO Strategy: The report identified poor performing landing pages which were being impacted by advert relevance issues, which informed our CX/CRO strategy, allowing us to prioritise landing page tests based on traffic and BERT Relevance score.