Gemini meta description generator

Given the text of a page, this returns a snippet-length description of it: a proposal for what a search engine displays beneath the title. The distinction from a summary is not stylistic. A summary addresses a reader who has committed. A description addresses one who has not, is scanning a list, and will spend under a second on the decision.

The operational bounds, since they constrain the claims this page can make: the family is a server generator, text is posted to our AI server, the allowance is 5 runs per rolling 24 hours, and input is bounded at 1 to 12,000 characters. The input is text, not a URL, because nothing in this route retrieves a page.

Most pages in this category present the output as the thing that will appear in search results. It is not. Google substitutes its own extract when it assesses the supplied description as weak, off topic, or a poorer match for the specific query than passage text available on the page. One of the four generator pages we fetched mentions this. Three do not.

The consequence is a different optimisation target. You are not writing the snippet, you are writing a candidate that must outperform an automatically extracted alternative, evaluated per query rather than per page. A description that is accurate and specific competes well against an extract. A description that is vague loses to one, which is precisely why vagueness is the expensive failure here rather than mere length.

Three steps, with the reasoning attached, because two of them look optional and are not.

Step 1

Paste the page's actual text rather than its topic. A description derived from the real content is the one least likely to lose to an extract drawn from that same content.

Step 2

Measure immediately. Character count is the only property of a description you can verify without guessing, so verify it before assessing anything subjective.

Step 3

Evaluate the first half in isolation. Truncation is not an edge case, it is the mobile default, so the leading fragment is the version most readers see.

Two of the fetched pages give ranges that do not overlap cleanly: 150 to 160 characters on one, 120 to 155 on another. Neither cites a source. Two further pages, including one of the largest tools in the category, state no length at all.

The contradiction is structural rather than sloppy. There is no character limit on a meta description. Search results truncate on rendered width, which depends on glyph widths, viewport and layout, so any character figure is a lossy proxy for a pixel measurement that varies by device. This is why a description of narrow characters survives longer than one of wide characters at identical counts. The defensible practice is a ceiling plus front-loading, and the ceiling this page uses is 160, because scripts/seo-audit.mjs in this repository sets DESCRIPTION_LIMIT to 160 and fails the build on any page that exceeds it.

No. The meta description is not a ranking input. Rewriting one does not change a position, and any argument for a description programme that rests on ranking improvement rests on a false premise.

The real mechanism is conversion of impressions you already earn. At a fixed position, the description is one of the few variables you control on the results page, and it competes directly against the descriptions immediately above and below it. That is a narrower claim than the category usually makes and it is the one that survives examination.

Detection is cheap and remediation is not, which is why most sites have the problem and few fix it.

Duplication is the default failure

Templates populate the field from a category name, and CMS fallbacks insert the site tagline. The typical finding is one description repeated across hundreds of URLs, which is functionally equivalent to having none.

Value concentrates at the top

A description converts impressions. Pages with no impressions gain nothing from a better one, so ordering the work by existing position, rather than alphabetically or by template, is the only allocation that pays.

This route is deliberately serial

No batch mode and no API, with 5 runs per rolling 24 hours. For a large site that is a multi-day effort, and stating it is more useful than implying a throughput the route does not have.

Input specificity determines output specificity

Pasting category boilerplate yields a description that fits every page in the category, reproducing the duplication problem through a different mechanism.

Four properties, ordered by how often they are absent in practice.

Leading identification

The first fragment states what the page is rather than what it covers. Since truncation removes the tail, any construction that defers the point loses it entirely on mobile.

Lexical overlap with the query

Present rather than stacked. Scanning is a matching task, and matched terms are where fixation occurs, which is a mechanical effect rather than a stylistic preference.

One discriminating specific

A figure, a constraint, a stated limit. Every competing result carries the adjectives, so only the specifics differentiate at the moment of choice.

No claim the page does not support

Overstatement converts an impression into a bounce, which is a worse outcome than no click, and it raises the probability of substitution by an extract.

Three contexts, distinguished by what makes the field a bottleneck.

Continuous publishing

Teams shipping weekly, where the description is filled last and worst and a template fallback is the realistic alternative to a generated draft.

Post-audit remediation

Where the missing and duplicate descriptions are already enumerated and the constraint is writing throughput rather than detection.

Cross-format authors

Writers whose article exists, often drafted in Gemini, and whose snippet does not. The two artifacts address different readers at different commitment levels.