Approve a research-led article when its evidence supports the claims that matter and its reader can use those claims to make the promised decision. Begin with the headline, the main finding and the action you expect someone to take. Then trace the important statements to the research. A bibliography can help you find sources. The review is where you establish what those sources actually prove and whether the article has made their limits clear.
Set aside separate time for factual review and editorial review. The fact checker should be able to say whether a source supports a statement about a date, a number or a business. The editor should be able to explain whether those facts answer the reader's question, whether the comparison is fair and what the article adds to existing coverage. A piece can pass one review and still need work in the other.
Disclosure. AI Visibility is published by Lunasol, which offers GEO and AEO services. This playbook uses public examples and a fictional approval exercise. No private client material or new live assistant observations were used. About AI Visibility
Read the promise before reading the footnotes
Ask the commissioning editor to state the reader's decision in one sentence. A reader might be deciding whether a cited source also supports a recommendation, whether an offer applies to them or whether a result justifies a change to a page. Compare that decision with the headline, opening and conclusion. If the headline promises current eligibility and the research contains only a historical promotion, the draft needs a narrower purpose or new evidence.
Read the nearest existing page. The new piece should make a material addition, such as a worked decision, original data or an evidence comparison the existing piece does not provide. The broader 90-day AI Visibility programme covers when research and content work might happen. An approval review has a narrower task. It decides whether the specific draft is accurate and useful enough to release. Google's people-first content guidance provides questions about original value, clear sourcing and whether readers learn enough to achieve their goal. Use them to test the draft's contribution. AI Visibility guides and 90-day plan · Google Search Central people-first guidance
Check what a person who arrives from search can do after reading the piece. Ask a colleague to read the draft without the brief and explain the decision in their own words. They should be able to find the material condition, tell where it applies and identify the next action. If they can only repeat the topic, ask the writer to add the missing decision or reorganise the evidence around it. This test is particularly useful when a technically accurate article leaves the practical conclusion buried in the middle.
Review the claims that could change the reader's choice
Have the researcher provide a claim record for material statements. Include the exact sentence, source and passage, date checked, scope, and any limitation. Open the cited source at the point where it supports the statement. A link to a company homepage may establish identity while providing little support for a claim about a particular service. A screenshot can show the value a reviewer saw on a date. A current availability claim needs current evidence appropriate to the service and location.
Check whether the sentence describes a direct observation, a publisher's statement, an inference or an example. Use wording that makes the distinction visible. If a business says it offers a service, attribute that to the business and retain any important qualifications. If the author inferred that two offers use the same terms, ask which source connects them. If a number belongs to a fictional example, mark it where the reader encounters it and keep it out of any findings summary.
Give a relevant subject matter owner the claims that depend on their expertise. GOV.UK's published fact-checking guidance distinguishes the content designer's responsibility for how a page is written from the subject matter expert's responsibility for factual accuracy. A commercial team can use that split. For example, ask an operations reviewer to confirm service conditions and ask the editor to check the explanation against the reader's decision. Bring factual amendments back to the editor before approving the final version. GOV.UK fact-checking guidance
Reconstruct one result from the underlying records
For a study, ask for the protocol and a sample of the original records. Look for the question wording, collection dates, surfaces, inclusion rules and handling of unavailable observations. Pick one important result and rebuild it. If the finding says a certain share of named businesses received a citation, establish what counted as a named business, which links counted as its own website and which answers entered the denominator. Check how repeated names were handled and whether the headline uses the same unit as the table.
The existing AI Visibility monthly reporting article describes a related-party study on Google AI Mode with a dated, bounded set of observations. It reports that 64 of the 78 businesses named in that set also had their own websites cited in the same answer. A reviewer can check that the rounded percentage follows from the stated numbers and that the sentence refers to the selected answers. That arithmetic check is narrower than verifying every underlying row or generalising to other assistants, sectors or periods. This playbook cites the public report's description and does not independently rerun its study. AI Visibility monthly reporting guide and study summary
Ask whether the text distinguishes missing observations from actual negatives. A planned answer that failed to load supplies different evidence from a captured answer that omitted the business. If the author changed a classification after reviewing the results, ask them to record the rule, which rows changed and whether the published numbers were recalculated. A reviewer who can reproduce one calculation and inspect the exceptions is better placed to decide where more checking is needed. A high-stakes or central result can warrant checking every contributing row.
Use the same discipline for a guide whose main evidence comes from documents. Take one recommendation and trace the steps from source to action. If a vendor's help page describes an export option, the article can report that description with the relevant plan or surface. A recommendation to buy the tool for a particular workflow also needs evidence about the fields the buyer can retrieve and the work those fields will support. Ask the writer to distinguish the documented feature from the test a prospective buyer would still need to run.
When two sources disagree, check whether they concern the same entity, period and terms before asking which is stronger. A contradiction may disappear once the dates or service areas are visible. If it remains, identify the person or source that can resolve it. Where the answer stays uncertain, the article can describe the disagreement if that helps the reader decide what to confirm. It should carry that uncertainty into any summary that repeats the claim.
Inspect the asset and its caption together
Treat an image or social post as a source with its own scope. Open the exact permalink and verify the account, date and material that supports the claim. Read any relevant image text, caption or timestamp. Then inspect the article's caption to see whether it describes what was actually checked. Record how the asset will be used and what permissions or publication method apply. A public permalink and permission to reproduce a file are different questions.
The official IKEA UAE Instagram post dated 5 November 2026 illustrates the split. Its image supplies a product price while its caption adds promotional timing. A reviewer looking only at one of those could miss a qualification that matters to a buyer. The merchant also publishes separate voucher terms that describe a different period and selected customers. Those terms should not be attached to the Instagram offer without evidence connecting the two. Use the specific source for the exact offer and keep current availability separate from the historical observation. Official IKEA UAE Instagram post · IKEA UAE voucher terms example
Check the same boundaries in the headline, callouts, metadata and any social excerpt prepared for the article. A careful body paragraph cannot correct an unsupported claim in a headline someone shares on its own. Verify the credit, publication date and commercial disclosure where readers will see them. If a source or embed becomes unavailable, decide whether a fallback link or a different asset can serve the same purpose. Hold the relevant claim if the asset was its only evidence.
Make an explicit approval decision
Imagine a fictional draft that helps UAE shoppers decide whether they can use a retailer's advertised offer. It contains an official post, a separate voucher page and a table that describes current availability. The editor can confirm that the brand published the dated post, while the claim of present availability needs a current check. The reviewer also finds that the two offer sources concern different terms. The remedy is to remove the unsupported connection and either narrow the article to a dated example or collect the current evidence needed for the original decision.
| Finding during review | Decision | Person who clears it |
|---|---|---|
| The body is accurate but the title promises a broader result | Revise the title and related metadata | Commissioning editor |
| A calculation uses scheduled runs as though all were observed | Rebuild the denominator and check every dependent statement | Research owner and factual reviewer |
| A current service condition is unconfirmed | Hold the claim or narrow the decision with approval | Service owner and editor |
| The source is valid but media reuse is unclear | Use an approved method or replace the asset and check the caption | Rights or publishing owner |
| The draft passes factual review but adds little beyond an existing guide | Revise the reader outcome or combine the useful material with the existing page | Commissioning editor |
Record the result so somebody else can tell what version was approved and what conditions apply. A brief form can travel with the package.
Research-led article approval
Topic identifier, draft version and reader decision
Material claims reviewed and evidence record
Calculation or sample checks performed
Asset provenance, use and disclosure check
Decision, required amendments and named approver
Recheck date or event that triggers review
If the result is revise, list the material issues and who should clear each one. Keep the decisions on each claim with the record. When the writer returns an amended draft, verify the change and its effect on nearby passages, the summary and metadata. If the article is approved, preserve the accepted version and link to its source record. Let the publishing owner confirm the final presentation and release authority as a separate step.
Before signing off, ask what evidence would make the approved wording stale. A promotion end date, a changed tool specification or a revised source can all trigger a check. An article that reports historical observations can retain them with their original dates. A recommendation that depends on today's conditions may need an earlier review. Approval should leave the next editor with enough information to tell the difference and update the right part.



