Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
CitePulse’s central insight is that a website can be easy for a machine to read yet rarely cited, cited accurately but infrequently, or visible in answers while still being difficult for a browser agent to use. Its audit keeps those outcomes separate—and reports “not determined” when access barriers or small samples make a score unjustified.
The figures below come from Lawrence’s maintainer-authored DEV Community case study, published September 24, 2026, describing CitePulse v1.7.0 runs on that date. They are observations from three anonymized targets, not independently replicated results or general benchmarks.
What does CitePulse audit?
CitePulse is presented as a local-first, open-source tool under the MIT license. Lawrence says the audited runs execute locally and that no data leaves the machine; those are claims in the case study, not independently verified operational guarantees. The article names Ollama and the local llama3.1:8b model for its reported run.
Recommended Free Tools
The audit treats “answer-layer” visibility as several related but different questions: can a machine access the site, does a cited page support the claim attributed to it, does the site appear in answers to tested prompts, and can a browser-driven agent interact with the site and complete a task? Its five principles are readability, citation support, retrieval in real prompts relative to competitors, agent task capability, and honest reporting when a measurement cannot be made.
The article describes nine KPIs spanning crawl accessibility, schema, llms.txt, citation correctness, citation rate, share of voice, interaction readiness, and task completion. The point is not that every site should maximize one composite score. Each KPI describes a different potential failure.
How to read the answer-visibility measures
Citation correctness is not citation rate
Citation correctness asks whether the cited page actually supports the statement made in the answer. Citation rate asks how often the target site appeared as a citation across the tested answers. A site can score well on correctness when cited, yet have a low citation rate because it is seldom selected. Conversely, frequent citation would not by itself establish that those citations support the claims.
Share of voice is relative to the tested prompt set
Raw and weighted share of voice describe visibility relative to competitors in the prompts tested. They are not estimates of market-wide visibility. Citation rate and share of voice are related but not interchangeable: one tracks appearances of the target as a citation, while the other expresses its relative visibility within the tested comparison.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsThese are proxy results, not tests of commercial answer engines
Lawrence says citation and share metrics are generated by a local model synthesizing live web-search results. He calls this “a proxy for AI-answer-engine behavior, not a live query to ChatGPT, Perplexity, Gemini, or Copilot.” Accordingly, the results should not be read as direct measurements of how those services answer users.
Interaction and completion measure browser actions
Interaction readiness and task completion concern whether a browser agent can act on a site, not whether a language model can describe it. A site might be discoverable in search results but difficult for an agent to navigate or use. These measures answer a different question from citation metrics.
What the three reported audits show
The case study reports three anonymized public-site audits. The table preserves the sample sizes because percentages based on small denominators can look more conclusive than they are. Every value is a reported case-study output from Lawrence’s DEV Community article, not an independently verified benchmark.
| Target | KPIs measured | Citation correctness | Citation rate | Raw share of voice | Weighted share of voice | Interaction readiness | Task completion |
|---|---|---|---|---|---|---|---|
| A, AI search-monitoring SaaS | 9 of 9 | 100.0% (N=10) | 55.6% (N=18) | 91.3% (N=18) | 89.1% (N=18) | 74.3% (N=35) | 33.3% (N=3) |
| B, European staffing and recruitment firm | 6 of 9 | Not determined: no citations to judge | 0.0% (N=18) | 0.0% (N=18) | 91.7% (N=18) | 85.7% (N=7) | Not determined: sample below the floor |
| C, cooperative bank | 5 of 9 | 100.0% (N=5) | 33.3% (N=18) | 86.5% (N=18) | 91.2% (N=18) | Not determined: authentication gated the probes | Not determined: authentication gated the probes |
The named figures are the article’s reported outputs: Target A had 100.0% citation correctness (N=10), 55.6% citation rate (N=18), 91.3% raw and 89.1% weighted share of voice (N=18 each), and 33.3% task completion (N=3). Target B had 0.0% citation rate and 0.0% raw share of voice (N=18 each), alongside 91.7% weighted share of voice (N=18). Target C had 100.0% citation correctness (N=5), 33.3% citation rate (N=18), and 86.5% raw and 91.2% weighted share of voice (N=18 each). Each number is from Lawrence’s 2026 CitePulse case study.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →Target A: accurate citations did not mean frequent citations or easy task completion
Lawrence reports that all 10 judgeable citations for Target A were supported by their cited pages. That 100.0% correctness result sits alongside a 55.6% citation rate across 18 answers and task completion of 33.3% across just three attempts. Read together, those figures distinguish reliable support when cited from frequency of appearance and from an agent’s ability to finish a task. The task sample is especially small, so the percentage is a description of those reported attempts, not a stable estimate of broader performance.
Target B: accessible to crawling, absent from tested citations
The article describes Target B as crawl-accessible but uncited in its tested prompt set: citation rate and raw share of voice were both 0.0% across 18 observations. Its weighted share of voice was reported as 91.7% across 18 observations. The contrast is a reminder that metric definitions and denominators matter; a strong value on one measure does not erase absence on another. Citation correctness was not determined because there were no citations to assess, and task completion was not determined because the sample fell below the stated floor.
Target C: high relative visibility did not ensure coverage of a basic query
For the cooperative-bank target, the article reports that only 6 of 18 answers cited the site, with coverage varying by query. It says the site was not cited for the basic identity prompt, “What is the bank?” Its reported 33.3% citation rate reflects those 18 answers; it does not mean every topic or query performed identically. Citation correctness was 100.0% among five judgeable citations, while authentication prevented interaction and task-completion probes.
Why a single average can mislead
A composite score can hide a load-bearing failure: crawl access cannot compensate for no citations in the tested prompts, and visibility cannot establish that a browser agent can complete a task. Lawrence puts it this way: “The verdict band is never the average of nine numbers; it is the report’s statement of the weakest load-bearing principle.” That framing is useful so long as readers also inspect the underlying measures, sample sizes, and conditions rather than treating the verdict band as a universal grade.
The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →- A crawl or schema pass establishes only what that probe checks; it does not establish that an answer engine will cite the site.
- Correctness evaluates support for citations that exist; rate evaluates how often citations appear.
- Share of voice is relative to a specific tested set, not the whole market.
- Interaction readiness and task completion describe browser-agent behavior, not answer visibility.
- “Not determined” is informative: it can mean no citations were available to judge, access was gated, or the sample was too small to justify a score.
How to compare audits responsibly
Two reports are meaningfully comparable only when their underlying test conditions and definitions align. A change in model, prompts, access, or sample composition may change the result without reflecting a real change in the site.
Best Value
- Match the test design. Use the same query set, prompt scheme, KPI definitions, model, and run dates wherever possible; record any difference.
- Record access conditions. Note crawl restrictions, authentication, challenge pages, and other barriers. The case study itself identifies a WAF challenge page returning HTTP 200 as a limitation of its crawl probe.
- Keep citation measures separate. Compare correctness only where citations can be judged, and compare citation rate with its own denominator.
- Interpret both share measures in context. Compare raw and weighted share using the same competitor set and prompts; neither is a market-wide statistic.
- Inspect agent outcomes and sample floors. Preserve interaction and task denominators, and leave values undetermined when access or sample size blocks a sound result.
- Track model and uncertainty. The article warns that historical runs using different local models may not be like-for-like. Without confidence intervals, score changes should not be treated as statistically significant.
What the case study can—and cannot—establish
These three anonymized examples demonstrate distinct failure profiles, but they do not establish population-level performance benchmarks. The samples are limited, and the article does not publish an independent population statistic. The results are also specific to the stated local-model and live-search proxy setup, rather than direct observations of ChatGPT, Perplexity, Gemini, or Copilot.
Lawrence discloses that he maintains CitePulse and says the three public-site targets were audited without prior arrangement; their identities are anonymized. The article identifies the project as CitePulse-public on GitHub, but the repository, license file, implementation, and audit manifests were not independently verified for this account. The case study’s MIT-license and local-first descriptions should therefore be understood as the project details reported by its maintainer.
Quick Recap
A practical checklist for reading a CitePulse report
- Check the run date, CitePulse version, model name and version, and whether the model is local.
- Identify exactly which prompts and competitors were included; do not generalize prompt-set share to the market.
- Read each KPI with its denominator and definition.
- Separate crawl accessibility, citation correctness, citation rate, share of voice, and browser-agent outcomes.
- Look for blocked or gated probes and retain “not determined” rather than converting missing evidence into a score.
- For trend comparisons, align test conditions and model versions, and do not treat changes without uncertainty estimates as significant.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
