New Feature: Grounding Snippet Ablation

Content Optimizer in the DEJAN app has a new Ablation tab, open to all users. It runs a prompt on Google, OpenAI or Claude, takes the sources the AI answer cited, and measures how many rank places each sentence of the top cited source is worth. In our first public example, one sentence of the source Google cited first accounts for its entire rank gain, and two other sentences pull it down.

Example: google algorithm volatility tracking tool

We ran the prompt "google algorithm volatility tracking tool" on Google from Australia, in our public demo project. You can open run 18 and check every number below.

Google cited five sources. The target was the one it cited first, an article on emarketeers.com. In the span test the article ranks 5th with its text removed and 4th with its full text.

One sentence is worth +1.04 ± 0.21 places, which is the whole gain:

Below are 5 algorithm monitoring tools that measure search engine volatility, so you can diagnose SEO issues almost in real-time.

Two spans pull the article down:

• "SEMRush Sensor - daily alert." (−0.36 ± 0.22 places)
• "Within SEMrush's paid tools, I particularly like its organic report," (−0.32 ± 0.18 places)

Every other sentence moved the rank by 0.27 places or less. The test took 356 rankings, and the whole run cost $3.65.

The run page shows a baseline rank of 3 for this target. That is the median of the three step 1 rankings. The 4th and 5th above come from the span test's own rankings, where each ranking is a single sample with the source order shuffled.

How it works

Step 1 runs the prompt, collects the text of each cited source, and ranks the unchanged set three times with the Content Optimizer ranker. The median is the baseline rank. Step 1 then cuts the target into spans and shows the exact cost of step 2. The default target is the source the platform cited first, and you can pick another.

Step 2 is the span test. It ranks the set many times with different combinations of the target's spans removed, and gives each span its Shapley value: the average number of rank places the target gains when that span is present. When the source names its brand, every mention is tested together as one span. The test stops when every ± range is 0.25 places or less, at the cost cap, or when every combination has been ranked.

We made the span test the default because in our tests, removing one thing at a time (a signal, the brand, a single sentence) did not move the rank. Measuring each span across many combinations did.

Platforms

• Google: the real snippet text the answer was grounded on (default), or a synthetic snippet.
• OpenAI: the real snippet text the model read.
• Claude: Claude does not return snippet text, so the app writes a synthetic extractive snippet from each fetched page.

Options

• Mode: Snippet mode ranks the snippets. Page mode fetches each cited page and ranks the full pages.
• Span unit: sentences or "..." parts in Snippet mode, paragraphs or sentences in Page mode.
• Cost cap: the span test stops before it spends more than the cap.
• Location: the country the prompt runs from, which changes the cited sources.
• Signal test: the older method, in Advanced options. A model lists the target's signals and the app removes each one once.

Where to find it

Open a project in the DEJAN app and go to Content Optimizer → Ablation. Past runs are listed under the form. On pay-as-you-go projects both steps are billed from the credit balance, and step 2 starts only when the balance covers its estimate.